Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
People's Republic of Bangladesh
Coastal Climate Resilience Infrastructure Project
(CCRIP)
Authors:
Aslihan Arslan
Daniel Higgins
Abu Hayat Md. Saiful Islam
The opinions expressed in this publication are those of the authors and do not necessarily represent those of the
International Fund for Agricultural Development (IFAD). The designations employed and the presentation of
material in this publication do not imply the expression of any opinion whatsoever on the part of IFAD concerning
the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its
frontiers or boundaries. The designations “developed” and “developing” countries are intended for statistical
convenience and do not necessarily express a judgement about the stage reached in the development process by a
particular country or area.
This publication or any part thereof may be reproduced without prior permission from IFAD, provided that the
publication or extract therefrom reproduced is attributed to IFAD and the title of this publication is stated in any
publication and that a copy thereof is sent to IFAD.
Arslan, A., Higgins, D. and Islam, A.H.M.S. 2019. Impact assessment report: Coastal Climate Resilience
Infrastructure Project (CCRIP), People's Republic of Bangladesh. IFAD: Rome, Italy.
Cover image: ©IFAD/GMB Akash
© IFAD 2019
All rights reserved.
1
Acknowledgements
The authors would like to thank all of those at the Local Government Engineering Department and
IFAD who assisted with the design and implementation of this impact assessment and provided
inputs for this report. This especially includes the CCRIP Project Director, Luthfur Rahman, and the
MEK Specialist, Shahjahan Miah, without whom this work would not have been possible, as well as
the GIS Specialist Neamul Ahsan Khan. From IFAD, we especially thank Peter Brückmann for his
extensive assistance during the data collection stage, David Hughes and Michelle Latham for their
extensive GIS support, and Philipp Baumgartner, Christa Ketting and Sherina Tabassum for their
support throughout.
Finally, we acknowledge the company hired to collect the data, BETS Consulting Services, along
with the sampled households themselves for their time and patience.
2
Table of Contents
Acknowledgements .......................................................................................................... 1
Executive summary .......................................................................................................... 3
Introduction ...................................................................................................................... 5
1. Project details, theory of change and main research questions .................................... 7
a. Project implementation ............................................................................................................... 7
b. CCRIP Theory of Change .......................................................................................................... 8
c. Research questions ..................................................................................................................... 12
2. Impact assessment design: Data and methodology .................................................... 13
a. Overall approach ........................................................................................................................ 13
b. Data ............................................................................................................................................... 14
c. Questionnaire and impact indicators ...................................................................................... 20
d. Impact estimation ....................................................................................................................... 25
3. Profile of the household questionnaire sample ........................................................... 29
a. Household characteristics by division ................................................................................... 29
b. Household characteristics by catchment area ...................................................................... 34
4. Results ........................................................................................................................ 36
a. Overall impacts of CCRIP ....................................................................................................... 36
b. Impact heterogeneity ................................................................................................................. 46
5. Conclusion .................................................................................................................. 51
References ...................................................................................................................... 53
Appendix I: Mean values for impact indicators ............................................................. 57
Appendix II: Distribution of Propensity Scores before and after trimming ................... 61
Appendix III: Results from the secondary IPWRA model ............................................. 62
3
Executive summary
The Coastal Climate Resilient Infrastructure Project (CCRIP) is a $150 million rural infrastructure
project which was implemented in 12 districts of Bangladesh since 2013, and is due to be completed
by the end of 2019. The project is funded by IFAD, the ADB, KfW of Germany, and the Government
of Bangladesh. The project aims to improve the connectivity of farms and households in the face of
climatic shocks, focusing on one of the most shock-prone areas of one of the most shock-prone
countries in the world. The main component of the project is the construction of improved markets
and market connecting roads, that are designed to remain useable during the monsoon season. This is
expected to improve sales of on-farm produce, along with access to inputs as well as opportunities for
off-farm income generation, leading to increased productivity and income. The project also aims to
improve women's empowerment by employing Labour Contracting Societies (LCS), consisting mainly
of destitute women, to carry out some of the construction work.
This impact assessment focuses on the activities funded by IFAD, which includes the strengthening of
markets and roads at the community and village levels. Using data from an in-depth household
questionnaire covering 3,000 treatment and control households, combined with extensive qualitative
interviews, we analyse the project's impact on a range of impact indicators relating to income; crop,
fish and livestock production and sales; assets, food security and education; financial inclusion; and
women's empowerment. We assess impact on the whole sample, as well as for a range of sub-groups,
including by geographic location, location within the market catchment area, and by livelihood
activity, integrating findings from the qualitative data to help to explain the mechanisms that shaped
the project's impact.
Regarding on-farm activities, we find that, despite a lack of impact on productivity, income from
selling crops and fish increased significantly (by 104 and 50 per cent, respectively). However, we do
not find a similar increase in income from the sale of livestock and livestock products. The lack of
impact on productivity was seemingly caused by persisting issues with accessing high-quality inputs
during the monsoon season, as well as households having limited capital to purchase these inputs.
Despite this, the project increased the amount of produce that was sold, the amount that was sold at a
market rather than from home or the farm gate, and increased the likelihood of growing cash crops,
leading to the large increase in on-farm income.
As well as improving on-farm income, the project also increased income from wage labour, which
together produced a positive impact on total income of 11 per cent, along with a four per cent
reduction in poverty. This increased prosperity was also reflected in reduced food insecurity and
increased ownership of households assets.
The project was intended to improve women's income generation and standing in the community,
mainly through its work with LCS, but we do not find an impact on women's contribution to
household income, or on their involvement in household decision making. When we analyse data
separately for Muslim and non-Muslim housheolds, we find that the project did improve these
indicators for non-Muslim households, suggesting that there were specific barriers faced by women
in Muslim households that the project was unable to overcome.
4
The additional sub-group analyses provided a number of additional insights. First, we find variation
in impact according to project district, which is caused by different impacts on income from crop
sales, livestock rearing, and wage employment. For the districts in Dhaka, for example, there was a
large positive impact on income of 21 per cent, which was driven by a large impact on livestock
income, which wasn't achieved in other areas. We also find that within the catchment areas of each
market, the larger impacts on income and poverty were achieved for the poorer, more remote
households. Finally, we find that the overall impact of the project was driven by improvements for
farming households (i.e. those with crop, fish or livestock production), while non-farming
households did not benefit significantly from the project.
Based on the findings of this impact assessment, we draw a number of important lessons for future
projects and policies that address the connectivity problem. First, future projects should pay special
attention to ensuring households have access to high-quality agricultural inputs, as improved access
to output markets does not necessarily improve access to inputs especially for credit constrained
households. This would help to increase productivity and thus stimulate larger impacts on income. In
addition to improving input access, agricultural productivity and income could also be improved by
providing complementary training and agricultural technology support.
Second, future projects should consider different components of beneficiaries' livelihoods and
provide activities to stimulate the main sources of income, which may vary for different areas. In
some cases for CCRIP, there was a lack of impact on the main income sources for some households
(mainly wage labour and household enterprises), leading to a lack of impact on total income. Impact
on income could thus be enhanced by offering complementary support for the livelihood activities
that are the most important in each local context. In the case of wage labour, this could involve a
redesign of the LCS activities to ensure that the valuable employment and training provided does not
remain short-term in nature and includes support to establish linkages with local labor market for
sustained impacts. In terms of household enterprises, these activities could be improved by
facilitating easier entry into local markets for small shops and traders, as well as providing credit and
training for setting up and managing these businesses.
Finally, future projects should provide more extensive support to improve the income generating
opportunities and overall empowerment of women, especially in countries such as Bangladesh where
women face ingrained barriers to their mobililty and autonomy. Based on the success of initiatives
such as BRAC's Empowerment and Livelihood for Adolescents program in Bangladesh (as well as
Afghanistan, Haiti, Sierra Leone, South Sudan, Tanzania and Uganda) future projects could provide
multi-faceted support to improve the hard and soft skills of women, provided within a safe space
environment, and involve the wider society to ensure sustainability of impacts.
5
Introduction
The Coastal Climate Resilient Infrastructure Project (CCRIP) is an inter-agency infrastructure
development intervention located in three divisions of southwest Bangladesh. The project
implementation started in 2013 and is due to be completed in 2019. It is funded with a combined
US$150 million from IFAD, the Asian Development Bank (ADB), KfW of Germany, and the
Government of Bangladesh, and is being implemented by a team from the Local Government
Engineering Department (LGED). In 2018, a combined data collection exercise was conducted by the
project team and a team from the Research and Impact Assessment (RIA) Division of IFAD to be used
for both the project's Mid-Term Report and an impact assessment study. This report details the design
and the findings of the impact assessment study.
CCRIP aims to improve the connectivity of farms and households in the face of climatic shocks,
focusing on one of the most shock-prone areas of one of the most shock-prone countries in the world
(Saha, 2014; Kreft, 2017). The project has three broad components: (i) Improved roads; (ii) Improved
market access; and (iii) Enhanced climate change adaptation capacity. One common theme across the
IFAD-funded components is their involvement of Labour Contracting Societies (LCS), which are
groups of mainly destitute women. These groups are contracted to carry out construction work, and
some of them are also provided with Women's Market Sections installed in community markets.
Upon completion, the project aims to reach 600,000 households from 32 Upazilas across 12 coastal
districts in the country. CCRIP was formed from the merging of IFAD's Sustainable Infrastructure for
Livelihoods Enhancement (SMILE) project with the ADB and KfW's Climate Resilient Infrastructure
Improvement in Coastal Zone Project (CRIICZP). This impact assessment focuses only on the impact
of the activities funded by IFAD, which consisted of the construction of climate resilient community,
union and village roads and markets, and the use and support of LCS.
In this impact assessment we test the impact of the project on a set of relevant impact indicators using
rigourous impact assessment methods, involving both quantitative and qualitative data and a carefully
constructed comparison group. This assessment has a number of benefits. Firstly, the insights from
this analysis help further understand how this type of project is expected to impact beneficiaries and
the contextual factors and barriers that can shape impact. Such insights can help to improve future
projects that seek to improve rural livelihoods in the face of climatic shocks, which are increasing and
are threatening rural poverty reduction worldwide (Kirtman et al., 2013; World Bank, 2017).
The second benefit of this analysis is that it contributes to IFAD's mandate to increase the
accountability of development spending. Along with 17 other projects, CCRIP was selected as part of
the IFAD10 Impact Assessment Initiative. Following on from the IFAD9 Impact Assessment
Initiative, the RIA division will use this set of impact assessments to extrapolate and estimate the
impact of IFAD's overall portfolio for its 10th replenishment period (2016-2018) (Garbero, 2016). This
initiative provides one of the most robust investigations into the impact of a development institution's
portfolio, and thus generates reliable insights into the results of IFAD's work and investments.
The final benefit of this work comes from its collaborative nature. By combining the work of the
project team and RIA, this assessment grants the opportunity to connect IFAD's field operations with
its impact assessment programme, creating a multi-stakeholder approach that facilitates insights that
6
are of the highest relevance and usefulness. In collecting detailed quantitative and qualitative data that
can be used to fulfil multiple reporting requirements, the work also provides an example of conducting
rigorous research on project performance in a cost-efficient manner.
The remainder of this report is structured as follows: Section 1 provides an overview of the context in
which the project was implemented, outlines the Theory of Change of the project, from which we
select our impact indicators, and presents the main research questions; Section 2 provides details of
the study methodology, the data, and the impact indicators analysed; Section 3 provides a profile of
the households included in the sample; Section 4 contains the results and discussion of the project's
impact; and Section 5 concludes with policy and programmatic implications.
7
1. Project details, theory of change and main research
questions
a. Project implementation
CCRIP effectively functions as three separate but conceptually linked sub-projects. IFAD's
component focuses on union and village roads and bridges, and on community and village markets;
while the ADB component focuses on larger scale Upazila roads, and large markets and growth
centers; and KfW focuses on the provision of cyclone shelters and other climate resilience support.
The IFAD interventions are being implemented in 32 Upazilas of 12 districts in southwest coastal
Bandladesh. Table 1 presents the CCRIP districts and the spread of project Upazilas across these
districts. These were identified from a set of 77 Upazilas that were assessed for inclusion using a
scoring system, which resulted mainly in the prioritisation of coastal, flood-prone, low-lying, and
infrastructure-poor chars. The scoring system was based on the following criteria:
Proportion of population below the
poverty line
Low wages for farm labour
Vulnerability to tidal surges,
storms, floods and river erosion
Remoteness
Poor communication (per cent of
paved road to total road)
Road density by population
Per cent of undeveloped markets
Table 1: Distribution of Upazilas across project districts
District Nr. Upazilas District Nr. Upazilas
Bagerhat 2 Khulna 3
Barisal 3 Madaripur 2
Bhola 3 Patuakhali 5
Borguna 4 Pirojpur 2
Gopalgonj 2 Satkhira 3
Jhalkati 1 Shariatpur 2
Within selected Upazilas, IFAD roads and markets are placed in areas that maximise the reach of
their benefits to poor people. This involves identifying the least developed unions and villages
within each Upazila, especially rural markets from char, low-lying, disaster-prone, and infrastructure
poor villages. For the LCS groups, households apply to be members, and are then selected based on
their poverty levels and experience in either construction or running a market stall.
8
For markets to be eligible for CCRIP support, they must meet the following criteria:
Strategically located and serve as an assembly market to benefit a large number of
villages and connect other larger market and growth centers;
Location not vulnerable to river erosion in the short and medium term;
Has potential for development in terms of availability of space and placing suitable
layouts;
Support from market stakeholders;
Agreement to share lease income with Market Management Committee.
Within the markets that meet this criteria, the final beneficiary markets are selected based on their
potential for poor women to participate in the construction of the market and as buyers and sellers;
the willingness of stakeholders to share part of the development cost to be used for the further
expansion of the works; and willingness of stakeholders to reserve sections for temporary sellers,
especially women and smallholders.
b. CCRIP Theory of Change
CCRIP is trying to solve a fundamental problem of rural development in southwest Bangladesh: the
low connectivity of smallholders' farms and households to markets, roads and urban centers (Rahman
and Rahman, 2015). The project has a particular focus on households living in char areas, which are
areas of land created by river sediment formed into sandbars along river channels, and are especially
remote and vulnerable to extreme weather shocks (Islam et al., 2014).
Low connectivity of households and farms hinders access to education, healthcare, financial and
support services, as well as employment oportunities. It also constrains access to input and output
markets, technology and productive facilities, and market information and extension services. This
lack of access has significant livelihood implications. At the household-level, limited access to these
services is widely regarded to negatively affect short and long-term livelihood quality and wellbeing,
including household food security and nutrition (Alkire and Santos, 2010,; Sibhatu et al., 2015;
Koppmair et al., 2017; Islam et al., 2018). At the farm-level, restricted access to input, technology,
extension and financial services can hinder the volume, quality and diversity of production, and
integration into value chains (Fan et al., 2012; Rehima et al., 2013; Bokelmann and Adamseged,
2016). Combined with poor access to vibrant markets and market information, plus high transport
costs, this can have a negative effect on the prices and profits that farmers receive for their goods
(FAO, 2003).
Both regular and unexpected climatic stresses exacerbate the connectivity issues in already remote
areas of southwest Bangladesh, especially in the char areas (Huq et al., 2015). During the annual rainy
season, many connecting roads become submerged and unusable, severely restricting transport. In
terms of unexpected shocks, the country experiences a tropical cyclone every three years, and a severe
flood every four-to-five years, with the southwest coastal region often bearing the brunt of the damage
(Nishat et al., 2013; Saha, 2014). For instance, two of the most recent major disasters in the country
damaged mainly the southwest region: Cyclone Sidr in 2007 caused 3,400 deaths and damaged
8,000km of roads; and Cyclone Aila in 2009 caused 180 deaths and damaged 7,000km of roads
(Relief Web, 2008; Relief Web, 2009).
9
CCRIP addresses the connectivity issue in the region by building and upgrading climate resilient roads
and markets. Roads are built to connect districts, villages and unions to each other and to markets.
These roads are made from materials that can withstand frequent submersion by salty or brackish
water. Roads are also raised and have higher and wider shoulders, with culverts and water gates
installed to manage flood water. Where suitable, vetiver grass is also used to line road slopes to
prevent erosion.
Households' market access is hindered by both a lack of transport infrastructure and a lack of physical
markets themselves. Cyclone Sidr alone is estimated to have caused damage and losses to the
country's agriculture sector of US$437 million, partly through damages to physical markets (Relief
Web, 2008). In order to complement the road work, CCRIP also establishes new markets and
upgrades existing ones. These markets range from "special" markets with over 200 permanent shops
serving over ten villages implemented by the ADB, to medium markets with around 100 permanent
shops serving up to ten villages, and smaller village markets with 10-50 shops serving up to four
villages implemented by IFAD. In terms of upgrades, CCRIP adds multi-purpose sheds, fish sheds,
boat landing platforms, open paved/raised areas, women's sections, toilet blocks, internal roads, and
improved drainage, depending on need.
The project also recognises that improving the management of markets in Bangladesh is key to market
sustainability (Ahmed, 2010). Market Management Committees (MMCs) are groups made up of
market users and local government, with a proportion of the committee having to be made up of
women, who are tasked with administration, maintenance and security of markets, but are often not
functional. As part of the market access component, CCRIP helps to organise these groups and
provides them with capacity building support. It also works with the local government to enforce the
legal stipulation that 25 per cent of the market lease income should go to the MMCs for maintenance
costs.
CCRIP is designed to improve the livelihoods of vulnerable women across the IFAD-funded
components, using LCSs as its primary tool. These groups consist of around 25 mainly destitute
women, who are trained and contracted to carry out road and market construction. In selected markets,
Women's Market Sections are also established. These areas are reserved for LCS members and
provide a permanent shop with favourable rent agreements in a safe environment. The project also
provides training to these groups to support other income generating activities. By offering these
opportunities, the project seeks to address the low social and economic status and the skills gap of
women in Bangladesh that restrict their livelihood activities (Roy et al., 2008).
Figure 1 presents the Theory of Change (ToC) for CCRIP, which maps the impact pathways expected
to link the activities of the project through outputs and outcomes to final intended impacts. The ToC
helps to identify the key indicators of success at each stage in order to track the expected impact
pathways of the project (White, 2009). In order for project activities to achieve their intended impacts,
there are a number of contextual factors that are required, which are outlined as part of the ToC.
Outlining these assumed conditions is an important part of the ToC that helps to identify additional
factors that need to be investigated in order to generate a thorough understanding of the project's
impact "story."
10
Figure 1: CCRIP Theory of Change
Women
Form and train Labour
Contracting Societies (LCS)
on construction and other
income generating activities
Contract LCS members to
conduct road and market
construction works.
Roads
Build and upgrade climate
resilient roads, bridges and
culverts
Markets
Build/improve climate
resilient physical markets
and their facilities
Restructure financial
management of markets
and build capacity of Market
Management Committees
Provide information on
farming practices, prices,
and weather through radio
service
Climate change
adaptation
Build/improve cyclone
shelters, upgrade access
tracks
Build community disaster
preparedness cap. building
Improved household
connectivity: schools,
hospitals, financial
services, support
services, etc.
Improved farm
connectivity: input and
output suppliers and
markets, technology/
facilities, other ag.
services such as
livestock vaccination.
Better managed, more
vibrant markets with
more buyers and sellers,
Sustainably structured
market management and
lease payment systems
More climate resilient
road and market
infrastructure
Increased employment
and income generating
capacity of women
Improved capacity to
rehabilitate infrastructure
after shocks
Improved access to
climatic shock protection
Higher education enrollment rates, reduced illness, better social security, etc.
Higher crop productivity and quality from improved input and financial service access
More diverse crop production from improved input and financial service access
Higher volume of goods sold and profits from sale from higher productivity and crop quality, reduced transport costs, value chain inclusion and better prices from better-functioning markets.
Livelihoods less affected by climate stresses and shocks
Diversified household income from improved income generating capacity of women
Increased bargaining power of women due to improved income generating capacity
Increased sustainably and smoothed income
Increased stability and resilience of livelihoods
Increased household and productive asset ownership
Increased food security
Empowerment of women
There is sufficient
demand and institutional
support for the activities
There are no issues with
acquiring land or other
materials for the work
Women are willing and
able to work in LCS
Roads, markets and
shelters are well placed
and well-designed
Training for LCS is
suitable
Farmers face no other barriers to their productivity or their market participation – lack of labour, lack of capital etc.
Income generating capacity is the only barrier to women's empowerment, they face no other barriers.
INPUTS AND ACTIVITIES OUTPUTS
OUTCOMES
IMPACTS
ASSUMPTIONS – Factors that need to be in place for the outputs, outcomes and impacts to be achieved
11
CCRIP is designed to increase household income through a number of intermediate outcomes. First is
increased market participation. With improved roads and accessible, better-managed markets, farmers
are expected to face lower costs associated with bringing their goods to market and thus to sell more.
Selling a higher proportion of their crops should lead to higher incomes, and over time an increase in
household and productive assets. The volume of sales as well as household food security are expected
to also be boosted by higher productivity. With better farm connectivity, and more sellers at markets,
farmers are expected to have better access to productivity-increasing inputs, technology, and extension
services. They may also be able to invest more in their production if improved connectivity leads to
improved access to credit providers.
Another expected intermediate outcome is higher prices. Better market access means more buyers at
markets, more demand, more options, which are likely to drive up prices. Although the upward effect
on prices of increased demand could be cancelled out by a downward effect of increased supply. Before
the project, farmers were often forced into taking lower prices by selling to traders directly after harvest
at the farm gate. With more favourable marketing options and improved access to inputs to improve
crop quality, better market information from CCRIP's radio service, and better connection to post-
harvest processing and storage facilities, this situation is expected to change (Barrett, 2008; Svensson
and Drott, 2010).
In addition to selling more and receiving higher prices, production and marketing expenditures are
expected to decrease, boosting profit margins and adding to the expected income effect. Along with
reduced transport costs, this is expected to occur as improved market access leads to improved input
access, which has the potential to increase the quality and profitability of farmers' crops (Gulati et al.,
2005; Khandker et al., 2009).
The project's work to build climate resilience is expected to ensure the intended outcomes outlined
above are not disrupted by climatic stresses and shocks, and to increase the overall stability and
sustainability of household livelihoods (Meybeck et al., 2012). By making roads and markets more
resistant to cyclones and floods, the vulnerability of livelihoods that are dependent on this infrastructure
is expected to decrease, meaning shocks have lower impacts, and households need less time and
resources to recover after them (Vallejo and Mullan, 2017). The cyclone protection and disaster
preparedness training of KfW and the support to MMCs to increase their capacity to repair markets
after a shock are also expected to contribute to improved climate resilience.
The above effects are targeted at all beneficiaries, whilst the LCS work is designed to produce income
and wellbeing benefits specifically for vulnerable women and women-headed households. In addition to
increased income generating capacity and increased economic opportunities from LCS participation,
LCS members' increased economic independence is expected to lead to the resource and power
allocation shifts needed for increased empowerment and wellbeing (Sheoran, 2016). Whether the LCS
member is the household head or not, members' households are also expected to benefit as LCS
members contribute more to household income and its diversification. Diversification potentially
further boosts the resilience of household livelihoods to shocks (Ellis, 1999; Arslan et al. 2018a).
At the bottom of Figure 1, we identify a number of key assumptions that are required to hold in order
for the above impacts to be fully achieved. If these assumptions do not hold, project's impact could be
constrained. For outputs to be achieved, it is assumed that the project will actually be able to identify
and acquire suitable land for roads, markets and shelters, and that these will be effectively placed, so
that they are used by intended beneficiaries. In terms of translating outputs into outcomes and impacts,
12
the project is designed under the assumption that households face no other significant barriers to their
production or market access such as a lack of capital. Finally, for the LCS activities, it is assumed that
women face no other barriers to their participation, and that women have demand for this support. Once
they have joined, it is also assumed that they face no other economic or social barriers to their
empowerment that the project does not address. The last assumption is a strong one in a country like
Bangladesh, where social norms constrain women in multiple ways, and the implications of this are
discussed further in the results and lessons learned sections below.
c. Research questions
This impact assessment focuses on the impact of IFAD's activities delivered through CCRIP as noted
above. The ToC diagram considers all of CCRIP's components in recognition of the expected overlaps
with the ADB and KfW work. Based on the expected impact pathways of IFAD's activities and the
potential complementarities with the activities of other agencies, this impact assessment answers the
following questions:
1. Did the community roads and markets delivered through CCRIP improve the household and farm
connectivity of beneficiaries? What were the subsequent effects on agricultural productivity, market
participation, and household income?
2. Did the IFAD activities delivered through CCRIP improve the climate resilience of beneficiary
livelihoods? What were the subsequent effects on household income levels and stability?
3. What were the impacts on women's livelihoods from the LCS-related activities? Were there barriers
to their participation in these groups? How effective were the different LCS activities (labour
contracting, income generation training,Women's Market Sections)?
4. What are the contextual factors that may have shaped the impacts of the project on beneficiary
households and women? What other lessons can be learned from the project that can be incorporated
into future rural development, climate resilience, and rural women's empowerment work in
Bangladesh?
13
2. Impact assessment design: Data and methodology
For this impact assessment a combination of quantitative and qualitative data was collected in order
to produce a holistic picture of the project's impact. The main data source is a quantitative household
survey of beneficiary and control households. This section presents the sample design of the
household survey and of the qualitative data collection, along with the statistical methodology
employed for impact analysis, followed by an overview of the key impact indicators used in the
analysis.
a. Overall approach
The CCRIP impact assessment is a collaborative effort between the project team and RIA. In order
for the quantitative household survey to be usable for both this impact assessmet and the project's
Mid-Term Report, we specifically sample beneficiary households who were reached during the early
stages of the project's implementation. In this way, data from these households can be used to
measure the project's performance against the mid-term indicators according to the log-frame in the
Project Design Report. At the same time these households have been exposed to the project's
activities for a sufficient amount of time, the data can also be used to measure the project's overall
impact.
The household survey covered both beneficiary and non-beneficiary households. The key to an
effective impact assessment is to compare a set of beneficiaries (the treatment group) with a set of
non-beneficiaries (the control group), who accurately represent how the set of beneficiaries would
have fared in the absence of the project. In this way, we are able to isolate the effect caused by the
project from other effects that occurred over time. The treatment population of interest for the
CCRIP impact assessment is all smallholder households within the catchment areas of CCRIP roads
and markets. The challenge of this assessment is therefore to identify a representative sample of this
population for the treatment group, and to identify a suitable comparable group of control
households.
To produce the final impact estimates, we conduct an econometric analysis of the household data,
comparing treatment and control households in a model that estimates the size of the effect on each
impact indicator, along with a measure of the effect's statistical significance. Statistical significance
represents the reliability of the result, giving the percentage probability that the result is a reflection
of reality and not due to chance (Gallo, 2016). In order to generate contextual insights, we conduct
our analysis on a range of sub-samples, in addition to the full sample. Table 2 presents the different
sub-samples that we test and the reason for assessing them.
14
Table 2: Overview of sub-sample tests
Sub-sample Hypotheses Tested
Households within 1km of market Did the impact differ based on market proximity?
Households within 2km of market but not 1km of
connecting road vs. Households within 2km of
market and 1km of connecting road
Did being close the connecting road as well as the
market provide a larger impact?
District Did district-specific local factors influence impact?
Households involved in farming activities (crop or
fish production or livestock rearing) vs.
Households not involved in farming activities.
Did project benefits on farm households also spread to
households primarily involved in wage labour,
household enterprises and other off-farm activities?
The type of impact assessment conducted in this report is termed "ex-post" as we use one round of
data collection without a baseline. Ideally the control group would be constructed at baseline, but
without this benefit we employ statistical matching techniques and expert consultations to improve
the accuracy of the treatment and control group comparison. This method has been shown to be
almost as effective as baseline control group construction in some cases (Dehejia and Wahba, 1999).
These matching techniques consist of a variety of matching algorithms to ensure that only similar
households are compared across the treatment and control groups, and is the primary method used
for ensuring accurate impact estimates in the absence of suitable baseline data (Austin, 2011).
b. Data
i. Sample distribution
The sample for the household survey is drawn from eight of the 12 project districts. Given the size of
the area, to collect data from all 12 districts would have been unfeasible and inefficient, thus we
selected eight districts covering the three project divisions (Barisal, Khulna, and Dhaka) according to
those with the largest CCRIP presence and those with the largest number of potential treatment and
control markets. In discussion with the project team, it was decided that the total sample size of the
household survey would be 3,000: a sample size deemed to provide sufficient power to detect
impact, and to cover a large enough area so that the estimation of impact is reliable and
representative.
In terms of the distribution of the sample, the sample frame was designed to achieve
representativeness at the divisional level using data on CCRIP investment by division as a proxy for
the number of beneficiaries in each division. Within each division, the sample is evenly distributed
across the eight districts. Table 3 presents the distribution of the sample across divisions and
districts, based on a sample size of 3,000, with 45 per cent allocated to treatment and 55 per cent to
control.
15
Table 3: Sample distribution across project divisions and districts
Division
Sample
allocation
(%)
Nr.
treatment
households
Nr. control
households
Nr households per
district
Barisal 61 824 1,006 Treatment = 206;
Control = 252
Khulna 16 216 264 Treatment = 108;
Control = 132
Dhaka 23 311 380 Treatment = 156;
Control = 190
The quantitative household data are complemented with qualitative data to contextualise findings
based on input from key informants and focus group participants. The qualitative data therefore
focuses upon the underlying impact mechanisms and the barriers to impact that may have been
faced. The collection of qualitative data was conducted simulatenously with the household survey
and consisted of the following:
Treatment markets: 5 x Key Informant Interviews (KII) with MMC members; 5 x Focus
Group Discussions (FGD) with road construction LCS members; 5 x FGD with market
construction LCS members.
Control markets: 3 x KII with MMC members or Market Manager; 2 x KII with Union
Parishad Womens' Representative.
Project staff: 3 x KII with senior project staff; 3 x KII with regional staff (one for each
project division).
ii. Identification of treatment and control groups
For both the treatment and control groups, we first identified suitable treatment and control markets,
with the intention of sampling households from villages within the catchment areas of these markets.
In order to capture the impact of being close to a CCRIP market, and the incremental impact of being
also close to a CCRIP connecting road, two catchment areas were defined for each market. To
capture the impact of being close to a CCRIP market without a CCRIP road, we sampled households
who are located within 2km of a CCRIP market (or control market) but not within 1km of a CCRIP
connecting road (or a un-improved connecting road in the case of control markets). To capture the
impact of being close to both a CCRIP market and a CCRIP connecting road, we sampled
households who are located within 2km of a CCRIP market and within 1km of a CCRIP connecting
road.
The radius size used for these catchment areas was decided with assistance from the project team.
They explained that these were distances within which households would travel to the market or
road, meaning these were the areas of expected impact. The 2km radius for the markets was also
used for a baseline study conducted of CCRIP. A small number of the control markets did not have a
main connecting road, in which case we sampled all households for the first catchment group.
16
Figure 2 maps the two catchment areas for one of the markets included in the sample (Pazakhali
Bazar). The orange circle represents the 2km radius around the market. The blue zone represents the
catchment area for household located withing 2km of the market and within 1km of the connecting
road. The other catchment area-for households located within 2km of the market but not 1km of the
road-consists of all other areas within the orange circle but not within the blue zone.
Figure 2: Diagram of the two sample catchment areas
Source: Authors' elaboration with technical support from IFAD-ICT Solutions Team.
The treatment markets for the sample were selected from CCRIP markets where market construction
work was completed before July 2015. The period 2013-2015 was considered as Phase 1 of CCRIP's
implementation, and taking households from the earliest phase means they will have been exposed to
CCRIP's work for a sufficient amount of time for impact to develop. Using households linked to
markets where work was completed later risks underestimating project impact by not allowing
enough time to pass before collecting the impact assessment data.
The control markets for the sample were selected from markets on CCRIP's back-up list, along with
additional markets identified by local project staff within the eight project districts covered by the
sample. The back-up list contains 19 markets located in the 8 project districts that were identified for
inclusion in the project according to the project's market selection criteria outlined in Section 1a,
however due to financial reasons these markets were eventually not covered. This means that
households linked to these markets should be suitably comparable to treatment households, and also
facilitates the availability of sufficient information and connections to identify a suitable sample.
Due to the limited number of markets on this list, a further 32 additional control markets were
identified based on expert consultations with the local project staff, who were instructed to identify
markets that would also have been eligible for inclusion in the project at the baseline stage.
Using the above strategies, we identified a total of 46 potential treatment markets for the sample, and
51 potential control markets. In order to identify the most suitable control markets we conducted a
scoring exercise, whereby we asked local project staff to assign scores for each market for a set of
Connecting
road
2km
1km
17
criteria that represented the selection criteria for CCRIP markets1. In this way we were able to build
a better picture of which markets would provide the most accurate counterfactual to treatment
markets. Through this process we eliminated 32 of the potential control markets which were deemed
as unsuitable for inclusion. This left us with 19 potential control markets.
In order to identify the final set of markets that were the most well-matched, we used GIS mapping
to obtain the population densities of the 2km catchment areas around each of the shortlisted
treatment and control markets. This allowed us, first, to identify whether any of the markets had
overlapping catchment areas, and to thus eliminate treatment and control markets that were
overlapping. Secondly, it provided an additional characteristic upon which to match treatment and
control markets—a characteristic that is linked to market size and the income level and
environmental quality of the surrounding area (Cropper and Griffiths, 1994). We therefore selected
the final set of treatment and control markets by producing matched pairs according to population
density and also, when possible, from within the same upazila to avoid the between upazila socio-
ecological hetorogeneity.
For each treatment and control market selected for the sample, the distribution of the sample across
the two catchment groups is equal in order to obtain a sufficient sample size to measure impact on
both of the catchment groups. From within the catchment areas of each market, households were
selected randomly. As we did not have lists of all households within each area from which to
conduct a purely random selection, we devised a three-stage process. Firstly, for each market, with
the assistance of the local leaders (primarily the Union Parishad Chairman), we compiled lists of all
of the villages located in each of the catchment areas, along with an estimate of the number of
households in each village. Our target was to collect data from a minimum of 15 households per
village, meaning the second stage involved randomly selecting the villages to be included in the
sample in order to meet this target. In cases where there was an insufficient number of villages in the
catchment area to meet the target, we included all villages located in the catchment area and
increased the sample size per village. For the final stage, within each village, households were then
randomly selected through the random walk method, using a sampling interval based on the
estimated village population size.2
Based on this strategy, the final set of treatment and control markets, and their associated sample
sizes are presented in Table 4 below, and Figure 3 contains a map of the sampled area where the
sampled markets and their catchment areas are plotted.
1 To reflect the CCRIP market eligibility criteria, scores were assigned for the following: (i) Based in char, low-lying,
remote, disaster-prone and infrastructure poor area (Yes/Somewhat/No) ; (ii) Connecting roads to market are dirt roads
that are not flood resistant (Yes/Somewhat/No); (iii) Market has a multi-purpose shed (Yes/No); (iv) Market has a fish
shed (Yes/No); (v) Market has a boat landing platform (Yes/No); (vi) Market has an open paved/raised area (Yes/No);
(vii) Market has a women's section (Yes/No); (viii) Market has an internal road (Yes/No); (ix) Market has improved
drainage (Yes/No). 2 For this method we first selected a landmark within the village (such as a school or communal area) and assigned four
teams to walk north, south, east and west of the landmark. The teams were assigned to sample households in their
direction according to a pre-assigned interval. This interval was calculated based on the total population of the village
and the required sample size for the village. We divided the total population by four (based on the four directions), and
the required sample size by four (based on the four directions and four teams), and then calculated the interval by
dividing the share of the total population by the share of the sample size. For a village with 100 households and a
required sample size of 16, the calculation would be as follows: (100/4)/(16/4) = 25/4 = 6.25. Meaning the sampling
interval would be every 6 households in a given direction.
18
Table 4: Distribution of final sample
Market name Division District Upazila Treatment
group
Sample
size
Joyer Hat
Barisal
Bhola
Burhanuddin
Treatment 103
Chanmiar Hat Control 126
Uttar Manika Bazar
Charfession
Treatment 103
Dolar Hat Control 126
Pazakhali Bazer
Patuakhali Sadar
Treatment 206
Akhai Bari Control 252
Gutia Bazar
Barisal Uzirpur
Treatment 206
Gondershor bazar Control 252
Badurtala Hat
Barguna
Pathorghata Treatment 206
Barotaleshwar Bazar Bamna Control 252
Chutukar Hat
Khulna
Bagerhat Sharankhola
Treatment 108
Rajapur Bazar Control 132
Ghorkumarpur Bazar
Satkhira Shyamnagar
Treatment 108
Patakhali Bazar Control 132
Suagram hat
Dhaka
Gopalgonj Kotalipara
Treatment 156
Hasua Bazar Control 190
Bairagir Bazar
Madaripur Rajoir
Treatment 156
Sonapara Bazar Control 190
Total 3,004
19
Figure 3: Map of treatment and control markets
Source: Authors' elaboration with technical support from IFAD-ICT Solutions Team. Each market is
represented by its centroid on the map and the circles around the centroid represent the cathment
area within 2km of each market.
Table 5 presents an overview of key descriptive statistics of the treatment and control groups, giving
an idea of the effectiveness of our sampling strategy in constructing comparable treatment and
control groups. Although in most cases the statistics are reasonably similar across the two groups,
there remains important statistically significant differences in household composition and religion,
and the average distance of the household to the sample market. This highlights the challenge of
creating accurate comparison groups ex-post, and underlines the need to use additional statistical
techniques as in this impact estimation analysis to minimise these remaining differences.
20
Table 5: Comparison of treatment and control samples
Treatment
(1,188 households)
Control
(1,552 households) Difference
Household size 4.57 4.53 0.04
Dependency ratio (%) 61.66 69.27 -7.61***
Age of h'hold head 49.82 48.31 1.51***
Average household age 31.56 30.51 1.05**
Religion (%):
Muslim
Hindu
Christian
83.75
14.81
1.43
94.85
5.15
0
-11.10***
9.66***
1.43***
Gender of household head (%):
Male
Female
90.15
9.85
86.21
13.79
3.94***
H'hold head is literate (%) 76.09 73.32 2.77*
Nr. literate h'hold members 2.02 1.92 0.10**
Nr. climatic shocks in the past year 0.15 0.16 -0.01
Distance to sample market (km) 1.21 1.09 0.12***
Note: *,** and *** indicate that results of a t-test for the difference between the treatment and
control groups, representing whether the difference is 90, 95 and 99% significant, respectively.
c. Questionnaire and impact indicators
The household questionnaire collected a wide range of information to create the impact indicators
and other variables used in the data analysis. The questionnaire was designed to collect detailed
information on agricultural production (separated by crop and by the three main cropping seasons),
fish production (separated by pond), livestock rearing, income from other sources, asset ownership,
food consumption, financial inclusion, shock exposure, social capital, access to services, and
household decision making. The questionnaire was conducted between late August and early
November 2018 and collected information covering the 12 month period between August 2017 and
July 2018.
In addition to the qualitative and quantitative data, the project team have also conducted the
following additional surveys: (i) Survey of LCS members; (ii) Child anthropometric measurement;
and (iii) Surveys of road and market performance, some of which are used to inform this study.
Based on the ToC for CCRIP presented in Figure 1, we use the statistical analysis outlined in the
proceeding section to test the project's impact on a wide set of indicators. The agricultural indicators
and those relating to fish production are all aggregated across plots and ponds to the household level.
21
All of the indicators used, along with how they are constructed and the impact areas they are linked
to are presented in Table 6.3
3 Appendix I contains the mean values for the treatment and control groups for all of these impact indicators.
22
Table 6: List of impact indicators
Indicator Description Impact area
Agricultural and fish production and sale
Gross value of agricultural/fish
production
Converts harvest of all crops/fish into a monetary unit. Equal to the income from crop/fish sales, plus the
value of non-sale uses, valued using the median price for the sample for each crop/fish when sold (Carletto
et al., 2007).
Volume of production; effectiveness/efficiency
of farming/fishing practices.
Gross margins/ha Equal to the gross value of production minus the value of all inputs. Inputs that were not purchased were
valued using the median price for the sample for each input. Effectiveness/efficiency of farming practices.
Proportion of harvest sold Gross value of the crops sold as a percentage of the total gross value of crop production. Market participation and access
Sold at a market Household sold at least a portion of its crop/fish harvest at a market, rather than from home or farm gate Market participation and access
Total revenue from crop/fish sales Cash income received from sale of all crops/fish. Market participation and access; income.
Nr crop varieites Count of the different crops grown. Input access, farming practices, shock
resilience.
Grew high value crops Household cultivated at least one high value seasonal or perennial crop Input access, farming practices
Land cultivated Number of hectares of land cultivated. Input access, wealth
Value of inputs (BDT/ha.) Calculated for agriculture and fishing. Total monetary value of all inputs used. Inputs that were not
purchased were valued using the median price for the sample for each input.
Input access; investment in agriculture;
effectiveness/efficiency of farming/fishing
practices.
Productivity of inputs (BDT) The gross value of production per one Taka of input used. Input access; effectiveness/efficiency of
farming practices.
Value of fish consumption
(BDT/capita) The gross value of fish that was produced and used for home consumption Market access and participation; nutrition
23
Table 6: List of impact indicators (cont'd)
Indicator Description Impact area
Livestock ownership and production
Gross value of livestock production Calculated in the same was as for crops and fish. Median price used to value livestock and
livestock products that were consumed at home or given away as gifts.
Livelihood practices, effectiveness/efficiency
of livestock prod.
Net value of livestock production Calculated in the same way as for crops and fish Livelihood practices, effectiveness/efficiency
of livestock prod.
Income from livestock activities Cash income from sale of whole livestock and livestock products, including cuts of meat, milk,
eggs and manure.
Effectiveness/efficiency of livestock prod.,
income.
Milk productivity (ltr/cow) Amount of milk produced per adult cow Input access; Effectiveness/efficiency of
livestock prod.
Livelihood composition and poverty
Gross household income per capita (BDT) Total cash income from all sources per household member Income, poverty.
Net income per capita (BDT) Equal to gross household income minus all economic expenditures Income, investment, livelihood effectiveness.
Above poverty line Household income is above the $1.90 per person per day poverty line set by the World Bank
(Ferreira et al., 2015) Poverty
Number of income sources Count of the number of different sources of income in the household Livelihood practises; livelihood resilience.
Income compositon Percentage of household income comprised from crop sales, livestock, formal and casual wage
labour, household enterprise, remittances, land rental and other (interest, pension, etc.) Livelihood practises, income
Sector participation A yes/no indicator for household being involved in the following sectors to earn income:
agriculture, fishing, livestock, waged labour, and household enterprise Livelihood practises
24
Table 6: List of impact indicators (cont'd)
Indicator Description Impact area
Assets, food security and education
Asset indices Composite indices constructed using Principle Component Analysis, calculated separately for
Productive assets and household assets (Filmer and Pritchett, 2001) Wealth, shock resilience.
Tropical Livestock Units Count of livestock owned, with different livestock converted into common unit (Jahnke, 1982). Wealth, shock resilience.
Food insecurity experience Standard indicators of food insecurity also adopted by SDGs (2.1.2), responses to eight questions
tallied into one Food Insecurity Experience Score (FIES) (Ballard et al., 2013). Wealth, food security
Dietary Diversity Score Score based on the consumption of different food groups in the past seven days (FAO, 2010). Nutrition
Childrens' school enrolment Proportion of school-age children currently enrolled in school Education
Financial inclusion
Have formal bank account or other
account
Yes/no indicators for whether the household has a formal bank account (held with a registered bank),
and whether they have another type of account, such as those held with a microfinance institution, a
non-bank financial institution, or a mobile money account (Anderson et al., 2016)
Wealth, market access
Received loan Yes/no indicator of whether household has received at least one loan in the past year Market access
Cash savings Total cash savings per capita, totalled across all savings locations. Market access, shock resilience, income
Women's empowerment
Womens' autonomous income
generation
Proportion of household income from the wage labour of female household memebrs or from
household enterprises owned or managed by female household members. Women's empowerment
Womens' decisionmaking involvment Female household members are involved (either individually or jointly) in decisions regarding:
household purchases, children's education, farm and livestock production and sale. Women's empowerment
25
d. Impact estimation
The main aim when estimating impact is to ensure that there is minimal difference between the
treatment and control group that are being compared. As well as our efforts to ensure that the two
groups are comparable during the sampling stage, as outlined above, we also employ statistical
techniques to further improve comparability during the data analysis stage. There are a number of
different statistical techniques that can be used to improve comparability and we use two separate
analytical models to test CCRIP's impact. Given the variety of approaches to ensuring comparability,
it is best-practice to employ one primary approach and one secondary approach that serves as a
means of testing the robustness of the results from the primary model. If the results are qualitatively
similar across the two approaches, then we can be assured that the results are valid and are not
model-specific.
Before running our impact estimation models, data quality checks were performed and 142 treatment
and 90 control households were removed from the final sample, either because the household was
incorrectly sampled from outside the 2km sampling zone around the focal market, or because their
data on key impact indicators (crop or fish harvest and income) were identified as outliers. In cases
where households had outlier data for minor variables, these values were replaced with imputed
values based on the distribution of the rest of the data.
Using the remaining sample, we then conducted an initial round of trimming using Propensity
Scores, whereby households that were clearly very different from the rest of the sample were
identified and dropped as they were unlikely to have comparable households in the opposite group
(Caliendo and Kopeinig, 2008). The propensity scores represent the likelihood of a household being
selected for project inclusion based on a set of relevant pre-project variables (or variables that are
unlikely to have been affected by the project), and are commonly used in impact assessment to
identify treatment and control units that would have been similar before project implementation, in
the absence of baseline data.4 Similar scores between a pair of treatment and control households
suggests that the two households were in a similar situation before the project began, meaning that
the control household can provide an accurate picture of how the treatment household would have
fared in the absence of the CCRIP.
Once we created these scores, we identified and dropped all households that had a Propensity Score
outside the area of common support, meaning that we dropped households with a score that was
above the highest score from the opposite group, and all those with a score below the lowest score in
the opposite group. For treatment households, the lowest score was 0.18 and the highest score was
0.88, and for control households, the lowest score was 0.17 and the highest score was 0.87. Based on
this we dropped ten treatment households and three control households.
In addition, we also dropped all treatment and control households that did not have at least one
household in the opposite group whose score was within 0.01 of their own score. From this, we
dropped an additional 13 treatment and 17 control households (see Appendix II for the distribution
of the propensity scores before and after these households were removed).
4 Specifically, the Propensity Scores in this case were created by running a logistic regression model where the dependent variable
is the binary treatment status (treatment or control) and the independent variables are variables linked to CCRIP selection and/or
livelihood capacity from 2014 (i.e. before the project was implemented), and then deriving the scores (representing the probability
of treatment) using the coefficients for each of the independent variables in the model, which represent their effect on the likelihood
of being treated (Caliendo and Kopeinig, 2008).
26
With the trimmed datset that consists of 1,192 treatment and 1,546 control households, we then run
our analytical models. The primary model we employ is a nearest neighbour matching (NN) model
(Khandker et al., 2010; Austin, 2011). This is similar to the approach used in the trimming process,
in that it creates Propensity Scores using the same set of matching variables, and identifies all
matched pairs within a set radius, although we narrow the radius in the model to 0.001 to increase
precision.5
The radius we have set for this model is relatively small, meaning that only the most well-matched
households are paired in the model. It is not always possible to set such a small radius size, however
because our treatment and control groups were already similar due to the efforts taken in the
sampling stage (as shown in Table 5), we were able to set this small radius with the confidence that
there are sufficient matches within the radius, and that we would not have to drop households that
did not have any matches within the radius.
One of the main benefits of the NN model is that it allows us to ensure that households are matched
exactly on variables of particular importance. As noted, we split the sample into two catchment areas
for each market: those 2km from the market but not from the connecting road, and those 2km from
the market and 1km from the connecting road. In order to ensure that we are not comparing
households located in different catchment areas, we implemented exact matching within each
catchment area as part of the NN model. Using GIS data, we also investigated implementing exact
matching on the distance of the household from the market (e.g above or below the median distance),
but we found that the level of comparability was best achieved when this distance was included in
the set of matching variables used to create the Propensity Scores, rather than using a categorised
version of this variable in exact matching.
Once the matched pairs are created using the NN model, impact is then estimated by taking the
average of the difference between the matched pairs. For example, if we compare the income per
capita of the treatment household with that of the control household within all of the matched pairs,
the impact estimation, termed as the Average Treatment Effect (ATE), is the average of the
differences in income across the pairs. The ATE is defined as:
Average Treatment Effect = E(y1i – y0i) (1)
Where y1i represents the outcome for treatment household i, and y0i represents the outcome for
control household i, and the E is the expectations operator
The secondary model we employ is an Inverse Probability Weighted Regression Adjustment
(IPWRA) model (Wooldrige, 2010; Austin and Stuart, 2015). In this model, impact is estimated
using a weighting rather than a matching approach. The model first assigns a weight to each
household in the analysis that represents the inverse of the probability of their receiving the
treatment that they actually received (beneficiary or control), with the probability calculated based
on the same set of variables used to create the Propensity Scores in the NN model. An econometric
regression model is then run with the weights applied, which estimates the average expected value of
the outcome if all units in the sample received the project and if all units in the sample did not,
5 A number of different radius sizes for this model were tested, and this radius produced the strongest set of
comparable matches whilst minimising the need to drop households who do not have a match (which is the downside
to using smaller radius sizes).
27
controlling for a set of relevant covariates, with the final impact estimate being calculated by
subtracting the control outcome estimate from the treatment estimate.
The set of control variables used in the IPWRA model cover household socio-economic
characteristics, land characteristics (for the agriculture indicators), geographic area, and exogenous
shock exposure. Control variables were chosen based on their likelihoods to have influenced the
outcome variable, while not having been affected by the project; thus different sets of control
variables were used depending on the outcome variable being analysed 6.
As mentioned, there is a wide variety of potential analytical models to employ for impact estimation,
and these must be selected based on which model provides the most reliable comparison between
treatment and control. In this case we tested a number of different models and found that the NN
model and IPWRA model were the most effective in this instance. The main way that we assessed
the effectiveness of the models was by assessing the change in the Standardised Mean Difference
(SMD) and in the Variance Ratio (VR) of the set of matching/weighting variables used for the
analysis (Austin, 2009).
The SMD measures the difference in the treatment and control means of a variable by the difference
in standard deviations, allowing for differences across variables to be compared in the same unit.
The VRs give a picture of how the relative variation across the treatment and control groups for each
matching variable has been altered. Table 7 contains the set of variables that were used in the
matching for the NN model and the weighting for the IPWRA model, and presents the change in the
SMD and the VR for each variable before and after the models are applied.
The average SMD between the treatment and control groups across the variables in the unadjusted
sample is 9.5 per cent, and we see that this is reduced to 2.1 per cent by the NN model and to 2.5 per
cent by the IPWRA model. For the average VR, we see that this has been reduced from 1.2 for the
raw sample, to 1.1 by the NN mode and to 1.1 by the IPWRA model. These statistics show that we
were able to significantly improve the comparability of the treatment and control groups for the
impact estimation with both models, highlighting that we can interpret our results with confidence
that there is negligible bias in the comparison. As the NN model performs slightly better in these
tests in addition to allowing for exact matching on catchment area, we employ the NN model as the
primary model for our analysis.
6 See Arslan et al. (2018b) for a detailed statistical specification of this model.
28
Table 7: Change in SMD and VR from the two impact estimation models
dd Standardised Mean Difference Variance Ratio
Variables Raw
(%)
After
NN (%)
After
IPWRA
(%)
Raw After
NN
After
IPWRA
HH size 1.83 -3.72 1.29 1.23 1.06 1.15
Dependency ratio -13.45 -3.13 2.64 0.73 0.84 0.89
Age of HHH 9.96 2.04 -2.82 0.99 0.97 0.98
Gender of HHH -11.66 0.45 1.50 0.76 1.01 1.04
Age * Ed of HHH 8.83 1.86 -4.19 0.94 0.95 0.88
Mean HH age 9.06 3.14 -1.86 0.99 0.97 0.96
Ed of HHH 7.00 1.28 -3.38 0.92 0.95 0.92
Ed*Gender of HHH 2.91 0.56 -2.64 0.82 0.91 0.92
Nr literate in HH 2.07 0.52 3.02 1.06 1.02 1.01
Nr English speakers in HH -0.84 -3.24 2.86 1.21 1.05 1.17
HHH speaks English -2.33 -0.26 1.08 1.07 1.01 0.97
Muslim -36.56 -4.39 1.77 2.78 1.13 0.97
Nr climate shocks -1.66 -0.91 -2.24 1.07 1.02 1.05
Productive asset index 2014 9.04 -4.48 1.78 1.92 1.33 1.86
Household asset index 2014 2.33 -2.46 3.26 1.12 0.96 1.16
Catchment area -28.01 0.00 0.65 1.20 1.00 1.00
Upazila 4.73 1.46 5.00 0.99 1.03 1.16
Distance to market 18.02 4.08 -3.42 2.44 1.66 2.40
Average (in absolute terms) 9.46 2.11 2.52 1.24 1.05 1.14
† These are interaction terms whereby two variables are multiplied together to incorporate their combined
effect (Baser, 2006).
29
3. Profile of the household questionnaire sample
a. Household characteristics by division
CCRIP is spread across three of Bangladesh's seven divisions: Barisal, Dhaka and Khulna. There is a
high level of diversity across these divisions that is likely to have an influence on the impact of the
project (Ahmed et al., 2013). For instance, as can be seen from Figure 3, the districts sampled for
Barisal and Khulna are located along the coastal areas and char zones, whilst the districts sampled
for Dhaka are located further inland. As noted, households living in the char zones face specific
challenges to their livelihoods, most salient of which is the more intense seasonal flooding in the
area, as well as being more remote (Sarker et al., 2003). There is also disparity in wealth, with Dhaka
the wealthiest of the three divisions, followed by Khulna (Ahmed et al., 2013).
Table 8 presents socio-economic statistics of the household sample for the three project divisions.
The divisions are similar in terms of average household size (around 4.5 members), average age of
the household head (around 49) and average age of household members (around 31). The gender of
the household head and the religion of households are similar in the Barisal and Khulna divisions—
with between 90-94 per cent of households having a male head, and between 93-96 percen belonging
to the Muslim faith, and the remainder belonging to the Hindu faith. Households in Dhaka district
are quite different in these characteristics, however, with a higher proportion of female-headed
households (around 23 per cent), a lower proportion of Muslims (76 per cent) and more Hindus (21
per cent) and Christians (3 per cent) compared to the other divisions. It is well-documented that
female-headed households face specific livelihood barriers, including higher poverty, lower access to
productive assets and capital, and constraining social norms, thus, we interpret the impact findings
with this background in mind (World Bank, 2001).
The households sampled from Dhaka are the wealthiest, whilst Barisal and Dhaka have similar,
lower levels of income. This wealth distribution is also reflected in the poverty levels of the three
divisions, with 82 per cent of households above the income poverty threshold in Dhaka, compared to
71 per cent in Barisal and 76 per cent in Khulna. The national poverty rate is reported to be around
15 per cent in the country,7 which indicates that CCRIP has effectively targeted areas with higher
levels of poverty and lower average incomes compared to the national average of BDT127,255 per
capita (around US$1,6048) (World Bank, 2017).
The project's targeting of households most exposed to climatic shock risk is also reflected in the
livelihood distribution statistics. In none of the divisions does a single income source provide more
than a third of total income, with households in all three regions showing signs of livelihood
diversification, which is a common form of ex-ante risk management in the face of persistent shock
threats (Jones et al., 2010).
Although the households sampled from the three divisions all have diversified livelihoods, there is
variation in how they diversify. In the Barisal division, around 18 per cent of household income is
7 http://povertydata.worldbank.org/poverty/country/BGD 8 Conversion made on 31 January 2019 from the following source:
https://www.reuters.com/finance/currencies/quote?srcAmt=127255&srcCurr=BDT&destAmt=&destCurr=USD
30
derived from the sale of crops, whilst 24 per cent of households do not cultivate any crops at all. In
Dhaka, crop sales provide 13 per cent of income and 34 per cent of households do not cultivate any
crops. In Khulna, just 5 per cent of income is from crop sales and 53 per cent of households do not
cultivate any crops. In this division, fish and livestock production play a more prominent role.
Across the divisions, formal wage labour and household enterprises are the major sources of income
for the households in the sample, although casual wage labour provides around 33 per cent of
income for households in Khulna, as expected given that housheolds in this division have the lowest
average income per capita. Household enterprises provide around 20 per cent of income in Barisal
and Khulna, and more than 30 per cent in Dhaka. The prominent role played by remittances is
evident everywhere, providing at least 10 per cent of total income, and they are especially important
in Dhaka providing around 27 per cent of household income, most likely coming mainly from those
working in the capital city. From these statistics it is clear that our impact analysis must focus on all
aspects of households' livelihoods, rather than simply agriculture, with improved access to waged
labour opportunities and demand for household enterprise services being key potential impact
mechanisms for CCRIP.
Although the contribution of wage employment to household income is high in all regions, there is
variation in the types of employment. In Barisal, only nine per cent of the formal employment is on
the farm, whilst 53 per cent is professional work (most commonly either office work or technical
jobs such as a mechanic), and 37 per cent is other off-farm work (such as construction). In Khulna,
professional work accounts for 86 per cent of formal employment, with nine percent constituted by
other off-farm work, and just five percent on-farm work. Finally for Dhaka, the largest share of
formal employment is provided by other off-farm work (44 per cent), with on-farm waged labour
accounting for a much larger share compared to the other regions (30 per cent).
Regarding household enterprises, in all divisions the majority (around 30-40 per cent) consist of non-
agricultural businesses such as small shops or services such as carpentry or barbering. In all districts,
taxi or pick-up services account for around 20 per cent of businesses, and in Dhaka, enterprises
offering professional services such as midwifery or tutoring are also common (around 30 per cent).
31
Table 8: Socio-demographic and livelihood characterisitcs of household sample by division
-1-
Barisal
-2-
Dhaka
-3-
Khulna
Household characteristics
Household size 4.56 4.55 4.46
Age of h'hold head 48.97 48.56 49.48
Average household age 31.08 30.37 31.30
Household dependency ratio (%) 63.27 77.65 60.75
Religion (%):
Muslim
Hindu
Christian
93.26
6.74
0.00
76.43
20.71
2.86
95.81
4.19
0.00
Gender of household head (%):
Male
Female
90.01
9.99
76.60
23.40
94.93
5.07
Per cent of adults who are literate
(%) 65.27 68.54 72.06
Per cent of school-age children in
school (%) 70.20 (1,408) 69.46 (505) 72.17 (371)
Livelihood characteristics
Total income per capita (BDT) 69,307.87 95,006.85 62,135.35
Above $1.90 per day poverty line
(%) 70.62 82.66 76.43
Land owned with title (ha.) 0.26 0.33 0.16
Proportion of income from (%):
Agriculture
Fish
Livestock
Formal wage labour
Casual wage labour
Household enterprise
Land rental
Remittances
Other (pension, etc)
17.60
1.30
9.09
24.76
8.37
21.22
1.03
11.79
4.29
13.40
0.67
5.64
14.06
1.92
30.25
2.25
27.40
4.38
5.24
4.17
4.30
17.43
32.98
22.78
0.52
9.03
3.08
Note: Unless otherwise stated in parentheses, the numbers of observations for these values are as
follows: Barisal = 1,692; Dhaka = 594; Khulna = 454.
32
Table 9 presents statistics on the agricultural practices of the households in the sample by division.
The more prominent role played by crop production in Barisal is reflected in the much larger harvest
value compared to the other two regions, both in terms of the volume produced, and the productivity
of production per hectare. Despite this, the largest average income from crop sales is for households
in Dhaka, followed by Barisal and Khulna, which is in line with the average total incomes across the
divisions.
The statistics for the dry and monsoon seasons relate only to the production of seasonal crops. In the
case of Barisal and Khulna, a larger proportion of households cultivated seasonal crops in the
monsoon season than in the dry season. Conversely, in Dhaka a larger proportion of households
cultivated seasonal crops in the dry season. This variation could potentially be linked to the types of
crops produced in Dhaka, as well as differences in seasonal demand for other income generating
activities (wage labor and household enterprises, both of which are more important in this division).
The importance of perennial crops is highlighted for Barisal in that a much smaller proportion of
crop production (in terms of value) is accounted for by seasonal crops, whilst seasonal crops provide
the main contribution in Dhaka and Khulna. The climate-related production issues in the project area
are reflected by the seasonal productivity statistics for the divisions, with total productivity higher in
all cases for the dry seasons, even in Dhaka where total production volume is higher during the dry
seasons.
In terms of income from crop sales, there are important differences in the selling practices of the
three divisions. In Barisal and Dhaka, around 70 per cent of crop-producing households sold some of
their harvest, while this figure is 50 per cent in Khulna. In terms of seasonal crops, in Barisal and
Dhaka the income per hectare from selling these crops is higher in the monsoon season. In total,
households in Dhaka sold the highest proportion of their crops (37 per cent), compared to 29 per cent
in Barisal and 22 per cent in Khulna. As is common for smallholders across the country, the majority
of households' harvest is dedicated to home consumption in all divisions (Anderson et al., 2016).
However, this is most prominent in Khulna (69 per cent of harvest), likely linked to the higher
poverty level in this division and also the higher production of rice, which is more likely to be
consumed at home than other crops (ibid). Households in Barisal lose the highest proportion of their
harvest to pests and shocks, although this percentage is low in all cases.
In terms of the types of crops that are grown, the main crop in all three divisions is rice, which is
grown by between 72 per cent and 85 per cent of the sample in the three divisions. In Barisal, beans
are an important crop, although not so in the other two divisions. Other divisional differences are the
importance of khesari in Khulna, and of coconuts and betel nut in Barisal.
33
Table 9: Agricultural production and marketing practices of household sample by division9
-1-
Barisal
-2-
Dhaka
-3-
Khulna
Total value of harvest (BDT)
Annual (incl. perennials)
Dry seasons
Monsoon season
933,839.50 (1,274)
23,508.84 (807)
22,188.45 (937)
54,899.21 (390)
52,820.32 (336)
18,129.56 (139)
34,478.86 (212)
14,707.32 (48)
28,407.84 (188)
Total value of harvest per ha.
(BDT/ha)
Annual (incl. perennials)
Dry seasons
Monsoon season
97,499.74 (1,274)
63,743.00 (807)
60,909.70 (937)
81,867.56 (390)
104,188.80 (336)
48,724.83 (139)
49,568.06 (212)
87,722.70 (48)
59,631.42 (188)
Total income from crop sales
(BDT)
Annual (incl. perennials)
Dry seasons
Monsoon season
60,073.53 (890)
18,904.40 (616)
27,266.45 (667)
75,782.55 (272)
39,581.44 (250)
51,603.67 (137)
28,104.85 (107)
14,508.33 (12)
27,498.93 (95)
Total income from crop sales per
ha. (BDT/ha)
Annual (incl. perennials)
Dry seasons
Monsoon season
34,071.75 (890)
44,689.18 (616)
62,481.47 (667)
87,590.46 (272)
63,191.68 (250)
163,600.20 (137)
26,418.81 (107)
50,061.43 (12)
38,874.06 (95)
Total area cultivated (ha.)
Annual (incl. perennials)
Dry seasons
Monsoon season
3.30 (1,274)
0.39 (867)
0.40 (937)
1.92 (390)
0.54 (337)
0.41 (142)
3.13 (212)
0.29 (68)
0.53 (188)
Crops grown (% of sample):
Rice
Beans
Khesari
Coconut
Betel nut
75.59
37.91
3.61
19.31
22.45
76.41
1.79
24.87
9.49
3.59
90.57
0.94
0.47
12.74
8.96
Harvest uses (% of total value):
Home consumption
Sold
Lost
Used for feed
Used for seed
Used for other purposes
60.47
28.95
1.19
0.20
1.15
8.04
52.35
36.50
0.73
0.13
1.43
8.86
68.89
22.17
0.37
0.66
1.35
6.56
Note: Unless otherwise stated in parentheses, the numbers of observations for these values are as
follows: Barisal = 1,274; Dhaka = 390; Khulna = 212.
9 Average statistics for annual, and seasonal indicators are compiled only for those households who cultivated during
each period. This is why the number of observation vary across indicators and why the annual average amounts are
not always above the seasonal average amounts.
34
b. Household characteristics by catchment area
Table 10 presents statistics of the sampled households separated by their location within the 2km
radius from the treatment or control market. The first column contains statistics for the full sample,
covering all those within a 2km radius of the market; the second is for all households located within
a 1km radius of the market, the third is for households located within 2km of the market but not
within 1km of the connecting road; and the fourth is for households located within 2km of the
market and also within 1km of the connecting road. For the treatment group, these different
groupings represent different levels of exposure CCRIP's support, with those located closer to the
market (Column 2) and those located closeer to both the market and the connecting road (Column 4)
having the highest level of exposure and thus being expected to receive the highest benefit.
From the table we see that the highest average income is amongst households located closer to the
connecting road, whilst there does not seem to be a difference in income for households according to
their distance from the market (based on statistics for columns 1 and 2). The lowest average income
is for households who are located within 2km of the market but not within 1km from the road,
suggesting that proximity to the connecting road is a key determinant of income variation. This
difference in wealth is not reflected by land ownership, however, with this figure averaging around
0.25 ha. for all groups.
One may expect that households located closer to the road would have higher school enrolment due
to better access, but this averages around 70 per cent in all cases. This may be due to the existence of
local schools in the area that do not require main connecting roads to access. The higher income of
people closest to the road is definitely highlighted, however, by the value of harvest, which is over
double the amount for some of the other groupings. Interestingly for this statistic, the value is not
higher for households located closer to the market.
In terms of income composition, households are similar across the different groupings, suggesting
that there is little variation in livelihood practices according to location in the catchment area. In all
cases, the majority of income is provided by household enterprises, followed by formal wage labour,
with income from crop sale providing around 13-16 per cent in all cases.
35
Table 10: Household and livelihood characteristics of the household sample by catchment area
-1-
2km from
market
-2-
1km from
market
-3-
2km from
market only
-4-
2km from market
and 1km from
road
Total income per capita (BDT) 73,690.67 73,045.28 69,638.25 78,528.26
Land owned with tenure (ha.) 0.26 0.25 0.26 0.26
School-age children in school (%) 70.36 (2,284) 69.59 (1,096) 69.75 (1,248) 71.10 (1,036)
Value of harvest per ha.
(BDT/ha) 88,833.39 (1,876) 48,939.31 (866) 46,025.22 (985) 136,157.80 (891)
Total area cultivated (ha.) 3.00 (1,876) 2.84 (866) 2.83 (985) 3.18 (891)
Proportion of income from (%)
Agriculture
Fish
Livestock
Formal wage labour
Casual wage labour
Household enterprise
Land rental
Remittances
Other (pension, etc)
14.64
1.64
7.55
21.23
11.05
23.44
1.21
14.72
4.11
13.12
1.22
7.34
20.24
11.90
24.62
1.11
16.33
3.83
13.76
1.57
6.54
22.68
12.79
23.01
1.17
13.92
3.98
15.69
1.73
8.76
19.50
8.97
23.94
1.26
15.68
4.26
Note: Unless otherwise stated in parentheses, the number of observations for each column are 2,740;
1,312; 1,491; and 1,249, respectively.
36
4. Results
In this section we present and discuss the results for CCRIP's impact on the set of indicators listed in
Table 6. We first present the results for CCRIP's impact according to households' proximity to the
CCRIP market, presenting results for the full sample from the 2km catchment area along with the
results for households located within 1km of the market. We then present results for CCRIP's impact
according by geographic location; by households' proximity to both the market and the connecting
road; and according to households' livelihood activities. The results we present are from the primary
NN model, whilst the results from the secondary IPWRA model are presented in Appendix III.
It is important to note that for the impact on specific sources of income (such as crop sales, fish
sales, or wage labour), the results apply only to those households, who were participating in these
activities. For instance, the impact we report on income from crop sales refers only to those
households who actually cultivated crops, and that for the impact on wage labour only applies to
those who actually worked as wage labourers during the study period.
a. Overall impacts of CCRIP
i. Agricultural production
Table 11 presents the impact of CCRIP on agricultural productivity and crop sale. For the whole
sample, we do not find a statistically significant impact on agricultural production for the 2km
catchment area for the whole 12 month cropping period. We find significant variation in impact by
season for seasonal crops, however, with a statistically significant increase in crop production of 94
per cent for the dry seasons (Kharif I and Rabi), but a negligible impact in the monsoon season
(Kharif II). In terms of gross margins, which is a measure of production efficiency, the project did
not have a significant impact for the full 2km catchment area. We also do not find that the project
increased the number of crop varieties grown by beneficiary households within the 2km catchment
area.
For the 2km catchment area, the results for crop sales differ from the crop production results. We
find that, for the full 12 month cropping period, income from crop sales increased by 104 per cent,
with a significant impact being achieved both in the dry and monsoon seasons—although the dry
season impact is slightly higher. We find that this impact is achieved despite the lack of a large
signficiant impact on agricultural production, partially due to a larger proportion of harvests being
sold amongst beneficiaries rather than being used for home consumption or lost to pests. In addition,
we find that beneficiary households were 11 per cent more likely to have sold their crops at a market,
rather than from home or at the farm gate, and were eight per cent more likely to have grown high-
value crops, impacts which also likely contributed to the total income effect.
For the 1km catchment area, there is evidence in some cases of an intensification of impact in terms
of crop production and sales. Although the impact on the value of crop harvest is still not significant,
we find a significant increase in gross margins of nine per cent, suggesting that production efficiency
was improved more for households located closer to the CCRIP markets. The effect on crop sale is
108 per cent for households located within the 1km radius as compared to the 104 per cent for those
in 2km radius, although the impact on seasonal crop income is not significant for the dry or monsoon
37
seasons, suggesting that impact on crop income came mainly through perennial crops for these
households.
The main implication of these findings is that improved market access for the sale of crops has
clearly been achieved through CCRIP, and has contributed to a very large increase in on-farm
income as a result. The statistically signficiant impact on crop sale income for seasonal crops in the
monsoon season is particularly salient with regards to this project, as it shows that, thanks to CCRIP
support, farmers were better able to sell their produce despite excessive rainfall which otherwise
damages roads and markets.
Insights from the qualitative data support this finding, with widespread reports that households were
able to increase their income because the markets were accessible and useable all year round, and the
markets were much better attended. In addition, the capacity building support provided to the MMCs
by CCRIP was reported to have greatly improved the sustainable management and maintenance of
the markets, with the increased attendance also providing higher income for the MMCs to perform
its duties.
The improved market access is also reflected by the findings that, although beneficiary households
were not growing significantly more than control households, their production was much more
market oriented, with improved selling practices. It may also have been the case that improved
market access did not stimulate increased production as households did not have to produce more to
account for crop losses caused by delays in accessing markets to sell produce. As a result,
beneficiary households achieved higher incomes from a similar amount of harvest compared to
control households. The fact that this has been achieved without a proportional increase in crop
production overall, combined with the finding that gross margins per hectare has improved only for
the sample within 1km of the market, suggests that benefits from improved input access due to
CCRIP's work may not have applied as much to more remote households compared to the crop sale
benefits.
38
Table 11: Impact of CCRIP on agricultural productivity and sales
2km from market 1km from market
ATE* Obs ATE Obs
Value of ag. production per ha:
Full year 14.2 1,876 9.5 866
Dry seasons only 93.9*** 1,876 16.3 866
Monsoon season only -13.0 1,876 15.8 866
Gross margins per ha. 2.0 1,876 8.7* 866
Proportion of harvest sold 4.8*** 1,876 2.9 866
Sold at market (%
likelihood) 11.0*** 1,269 15.2*** 598
Income from crop sale per ha.:
Full year 104.0*** 1,876 108.3*** 866
Dry seasons 130.2*** 1,876 67.6 866
Monsoon season 69.8** 1,876 65.6 866
Nr. crop varieties (count) -0.02 1,876 -0.1 866
Grew at least one high-value
crop (% likelihood) 7.6*** 1,876 1.3 866
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels,
respectively.
*Coefficients represent percentage change unless specified otherwise.
Table 12 presents the impact of CCRIP on agricultural input use. For the 2km catchment area, we do
not find any significant impacts on the volume of inputs used, including for the amount of land
cultivated, or the amount of seeds, fertiliser, crop protection or labour used during the 12 month
cropping period. As shown in Table 12, we find that gross margins were not significantly improved
amongst beneficiary households for the 2km area, and this is reflected by a lack of a significant impact
on the input productivity (the amount of output produced by one unit of input) of fertiliser, crop
protection, and labour. Only in the case of seed use do we find a positive and signficiant effect on
efficiency, with an increase of 11 per cent in the units of output produced by a unit of input.
As with the results from Table 11, the project's impact on input use intensifies in some cases when we
focus on households located within 1km of the market. Amongst these households, we find that land
use increased significantly, by around 1 hectare, as did the amount of fertiliser and crop protection that
was used (by 53 per cent and 65 per cent respectively). In addition, the input productivity of seeds
increased significantly by 16 per cent, as opposed to 11 per cent for the 2km area. However, as with the
2km area, there was still no significant impact on the productivity of fertiliser, crop protection or labour.
CCRIP was expected to improve households' access to agicultural inputs. In terms of the amount of
inputs that were used, this was not significantly increased for the 12 month period, something which
may have contributed to the lack of a significant impact on crop production. Based on the slightly more
favourable impacts for the 1km catchment area, this lack of impact particularly applied to more remote
households. It was widely reported in the qualitative data that CCRIP markets improved access to
agricultural inputs, but that households often lacked capital to invest in these inputs. Based on this, it
39
may have been the case that physical access to inputs was improved by the project, but additional
barriers such as a lack of capital served to curtail impact in this area.
sThrough additional seasonal analysis, we find that input use was significantly increased for seasonal
crops during the dry season, which again can be linked to the finding for crop production for this season
(where we find a significant positive effect). This additional insight suggests that a lack of impact on
the 12 month period is driven by a lack of input access during the monsoon season. Whilst we find that
crop sale was significantly increased during this season—suggesting the project was able to solve the
barriers to output market access posed by the rains during this season—it is possible that it was not able
to solve the barriers to input market access during this season. Although the positive effect on seed
productivity suggests that households may have been able to access better seeds, but not other inputs
Table 12: Impact of CCRIP on agricultural input use
2km from market 1km from market
ATE* Obs ATE Obs
Land cultivated (ha.) 0.5 1,876 0.9* 866
Value of inputs per ha. 12.9 1,876 19.0 866
Seeds 4.8 1,876 20.4 866
Fertiliser 10.1 1,876 53.2* 866
Crop Protection† 8.2 1,876 64.6** 866
Labour 12.5 1,876 33.6 866
Input productivity of:
Seeds 10.9** 1,610 16.1** 750
Fertiliser 4.4 1,525 -11.2 708
Crop Protection 9.2 1,409 7.0 636
Labour 5.9 1,876 -2.6 866
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
† Crop protection includes following inputs: pesticides, herbicides, fungicides, insecticide
ii. Fish and livestock production
Table 13 presents the impact of CCRIP on fish production and sales (for those who produced fish
during the study period), which bare some similarity to the project's impact on crop production and
sale. We find that the value of production for fish producers in the 2km sample was not increased,
but as with crop sales, we find a large positive impact on income from fish sales of 50 per cent. We
do not find a positive impact on the likelihood of fish being sold at market rather than from home or
at the farm gate, which may be due to the small proportion of fish-producing households who
actually sold fish (18 per cent of fish producers). Likely due to the lack of impact on fish production,
we do not find an impact on the amount of fish harvest used for home consumption. We find similar
results in all cases for the 1km sample.
As with crop sales, the results for fish production highlight the effectivenss of CCRIP in improving
market access for selling goods. As well as potentially being due to a lack of appropriate inputs, the
lack of impact on fish production may have also been due to households not having to produce more
40
to account for losses of their fish harvest due to delays in accesing markets to sell them, an impact
that has also been identified for similar IFAD-funded projects (Brett, 2019). One difference from the
project's impact on crop production, however, is that the impacts on fish production and sales do not
differ by distance to the market, suggesting there are specific barriers to crop production and sales
for more remote households that do not apply to fish.
Table 13: Impact of CCRIP on fish production and sale
2km from market 1km from market
ATE* Obs ATE Obs
Gross value of fish production -17.8% 1,526 -2.8 708
Net value of fish inputs -6.6** 1,526 -10.4** 708
Revenue from fish sales 49.5** 1,526 52.5* 708
Sold fish at market (% likelihood) 10.5 272 -8.0 111
Value of fish consumption per
capita -6.2 1,526 -18.8 708
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
Table 14 presents the impact of CCRIP on livestock production and sales (for those involved in
livestock production as an income source). We do not find a positive impact on the value of production
of livestock for livestock producers, or on income from the sales of livestock and livestock products.
However, we do find a significant 62 per cent increase in the milk productivity of cows. This could
potentially be due to beneficiaries having better access to inputs that could improve cow's milk
productivity, or access to more productive calves and cows themselves. In addition to these results, we
also find that households in the 2km sample were more likely to have consumed meat in the past seven
days.
These results suggest that the impact of improved markets and roads on the sales of crop and fish
produce did not apply to livestock, and thus did not serve to enhance livestock rearing as a livelihood
option. Livestock are usually traded at dedicated markets and livestock trading at the community
markets that were improved by CCRIP is not very common as they are usually taken to larger markets
in more distant locations. In the qualitative data, a common request for future projects was training and
other support for livestock rearing, suggesting that low capacity amongst beneficiaries for livestock
rearing may have also hindered the project's impact in this area.
41
Table 14: Impact of CCRIP on livestock rearing and sale
2km from market 1km from market
ATE* Obs ATE Obs
Gross value of livestock production 10.2 2,171 15.8 1,006
Net value of livestock production 0.9 2,171 1.3 1,006
Revenue from livestock and product sales 12.4 2,171 19.9 1,006
Sold livestock or livestock product (%
likelihood) 1.2 2,171 2.0 1,006
Milk productivity per cow 62.0* 888 -9.4 404
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
iv. Total income and livelihood composition
Table 15 presents the impact of CCRIP on household income and livelihood composition. For the
2km catchment area, we find that CCRIP significantly increased household's total income by 11 per
cent. In addition, beneficiary households were found to be four per cent more likely to be above the
poverty line. We also find that income for wage workers increased by 18 per cent for the 2km
catchment area. Despite these improvements, we do not see an impact of the project on net income,
which may be due to households investing more in their long term production capacities increasing
their expenditures. We also do not find an impact on income from household enterprises or the
number of income sources in the 2km sample, indicating that income diversification did not increase.
Looking at the composition of household income in the 2km catchment area, we see that the
contribution from fish sales was the only income source whose proportion increased significantly,
rising by around 1 percentage point. The only other significant change in income composition is a
significant reduction in remittances (by 2 percentage points), thus suggesting that beneficiary
households maintained their income generating portfolio while increasing total income, hence
needing less remittances.
For households within the 1km catchment area, incomes became more concentrated around crop and
fish sales, reflecting the large impact on these sources. Despite these increases, we see a lower
impact of the project on total household income compared to the 2km sample due to a significant 4
percentage point reduction in the contribution of income from household enterprises, and a
significant 0.8 percentage point reduction in land rental income. In terms of both livelihood diversity
and poverty reduction, however, the results for this sample are more positive than for the 2km
sample. Households within 1km of CCRIP markets are around 5 per cent less likely to be poor, and
there was a slight but significant increase (averaging 0.3) in the number of income sources.
For the 1km catchment area, it is perhaps surprising that total income did not significantly increase
based on the positive impacts on crop and fish sale income for these households. This is likely due to
the fact that crop and fish sales account for just 14 per cent of total household income for these
households, whilst waged labour and household enteprises account for around 38 per cent (see
Section 3b, Table 10). The benefits for these households from the project therefore applied to a more
minor area of their livelihoods, whilst the primary areas were not significantly improved, thus
leading to a curtailed impact on total income.
42
Table 15: Impact of CCRIP on income and livelihood composition
2km from market 1km from market
ATE* Obs ATE Obs
Gross income per capita 10.5** 2,740 9.0 1,312
Net income per capita 0.01 2,740 -1.02 1,312
Income from waged labour
per capita 18.3*** 1,291 11.2 607
Income from household
enterprise per capita -9.1 847 12.7 423
Nr. income sources (count) 0.1 2,740 0.3** 1,312
Above $1.90/ day pov. line
(% likelihood) 4.4** 2,740 5.3* 1,312
Proportion of income from
(percentage points):
Crop sale 1.4 2,740 5.4*** 1,312
Fish sale 0.9** 2,740 1.4** 1,312
Livestock sale -1.4 2,740 -0.9 1,312
Formal waged labour 2.3 2,740 1.2 1,312
Casual waged labour -0.6 2,740 -1.2 1,312
Household enterprises -0.7 2,740 -4.3* 1,312
Remittances -2.0* 2,740 -2.1 1,312
Land rental -0.4 2,740 -0.8** 1,312
Other sources -0.2 2,740 0.6 1,312
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
v. Assets, food security and education
Table 16 presents the impact of CCRIP on household assets, food security and education indicators.
We find that household durable asset ownership was significantly increased by the project in both
catchment areas (the impact is not significant for productive assets only). The impact for ownership
of livestock measured in TLU is not significant for the 2km sample, but slightly negative and
significant for the 1km sample. Regarding food insecurity experience indicators, we find that the
project significantly decreased food insecurity experience both using the full FIES score and
components. Beneficiaries were at least 10 per cent less likely to have worried about having enough
food, and to have been unable to eat healthy food due to a lack of resources during the past year.
Impacts on dietary diversity score and school enrolment are not statistically significant for the 2km
sample.
As with other areas, we find some evidence of more favourable impacts for the 1km sample. For
these households, there was a larger significant increase in the ownership of household assets and a
larger reduction in food insecurity concerns. Although there was also a small and significant
decrease in the dietary diversity score, while the impact on proportion of school-age children
enrolled in school is not significant.
43
The increase in household assets suggests that households may be purchasing assets as a form of
saving and risk mitigation, as opposed to keeping cash savings, which we found was not improved
for the 2km sample. This potentially has positive implications for these households, whose livlihoods
are plagued by a variety of shocks. This finding – combined with the finding of increased income
diversification for the 1km sample – suggests that, by improving the income of beneficiaries, the
project has improved the resilience of households, which has positive implications for the long-term
sustainability of the project impacts.
The contrasting results of the project on food security and dietary diversity (a proxy for nutrition)
may be explained by the high prevalence of poverty within the sample as highlighted in Section 3.
Concerns about having enough food is an issue that applies to the poorest households, who lack
sufficient income to provide for their basic needs, whilst dietary diversity is the concern that follows
once a household has achieved their basic requirements for sustenance. Based on this, the results
suggest that the project was able to help poor households to better meet their basic food needs, but
had less success in helping relatively wealthier households, who were already food secure, to
enhance their dietary diversity.
Table 16: Impact of CCRIP on assets and food security
2km from market 1km from market
ATE Obs ATE Obs
Asset indices:
Household durable
assets 0.1** 2,740 0.1** 1,312
Productive assets -0.02 2,740 0.04 1,312
TLU -0.1 2,740 -0.1* 1,312
FIES score -13.1%*** 2,740 -18.0%*** 1,312
Worried about having
enough food (% likelihood) -9.9*** 2,740 -14.8*** 1,312
Unable to eat healthy food
(% likelihood) -10.7*** 2,740 -13.6*** 1,312
Dietary Diversity Score -0.1 2,740 -0.5*** 1,312
School-age children enrolled
(percentage points) -0.4 2,204 1.8 1,056
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1%
levels, respectively
vi. Financial inclusion
Table 17 presents CCRIP's impact on indicators of financial inclusion. Smallholder farmers across
the world face substantial barriers to obtaining formal bank accounts for payments or savings, as
well as accessing loans (Anderson et al., 2016). Only around 21 per cent of households held a
formal account at the time of the questionnaire, and 20 per cent held another type of account.
Although the impact on the likelihood of having a formal bank account for the 2km sample is not
significant, for the 1km sample we find that the project improved this likelihood by around six per
cent. We also find that the likelihood of having an account other than a formal bank account has
decreased by around seven per cent for the 2km sample.
44
The lack of impact on financial inclusion for the 2km sample is perhaps explained by the lack of
direct focus by the project on improving financial inclusion. As shown in the Theory of Change,
CCRIP's direct impacts were expected through improved market access, and although the results
show that this indeed was improved in terms of buying and selling goods, it does not seem to have
applied to accessing financial services such as bank accounts, savings and loans. It may perhaps be
the case that the institutions that provide these services are based further away than the community
markets that were improved by CCRIP, which we cannot test with our data. Nevertheless, a lack of
access to credit was cited as a key barrier to improving livelihoods in the qualitative data, suggesting
that future projects may benefit from targeting this constraint.
Improved financial inclusion and savings may have also been expected to improve as an indirect
result of higher income stimulated by the project. As we have seen above, CCRIP had a significant
impact on household asset ownership, which is also a common form of saving amongst smallholders
in the country (Anderson et al., 2016). These findings together suggest that households may indeed
be saving more as a result of the income improvements induced by CCRIP, but this is in the form of
assets rather than cash, potentially as a result of persisting issues of access to formal financial
services.
Table 17: Impact of CCRIP on financial inclusion
2km from market 1km from market
ATE* Obs ATE Obs
Have formal bank account
(% likelihood) 2.8 2,740 5.7** 1,312
Have other account
(MFI, NBFI, mobile money)
(% likelihood)
-6.7*** 2,740 -0.8 1,312
Took a loan in past year (%
likelihood) -1.03 2,740 -5.3 1,312
Savings per capita -6.1 2,740 18.6 1,312
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
vii. Women's empowerment
Table 18 presents the impact of CCRIP on indicators of women's autonomous income generation and
their involvement in household decision making. We find that, regardless of the catchment area, the
project did not have a significant effect on the proportion of household income contributed by
women's activities, or on the likelihood of a female household member owning their own enterprise.
We also do not find a significant impact on women's involvement in household decision making
regarding household purchases, children's education or agricultural production or marketing.
Regarding decision making on education, around 83 per cent of the treatment sample reported that
female household members were involved either individually or jointly with male members,
suggesting that there was limited room for improvement in this area. However, there was much-
needed improvement in the other areas, with only 40 per cent of households having women involved
in household purchase decisions, and 58 per cent in terms of agricultural production. In addition, an
45
average of only two per cent of household income is contributed by women's autonomous activities,
and only two per cent of treatment households have a woman owning her own enterprise.
Overall there is little evidence from these results that CCRIP's market and road infrastructure
strengthening helped to improve women's empowerment in terms of autonomous income generation
and household decision making. This may be because these activities of CCRIP did not specifically
target women. The parts of the project that did target women were its use of LCS and the
establishment of women's market areas. As these activities were not expected to reach as many
households as the market and road infrastructure support, it was not feasible to design the household
survey sample to measure the impact of these activities, therefore there may have been an impact of
the project on women's empowerment that is not detected by this assessment. From the qualitative
data, it was reported that the LCS provided a valuable means of gaining income and training for its
members, and served to improve women's standing in the community. However, it was also reported
that some women were forbidden from joining the LCS by their husbands, and that after the work
with CCRIP had finished, female members had difficulty in obtaining additional employment, and
when they did find work their wages were often lower than men's. An additional survey was
conducted that focused on LCS members which may detect further an impact in this area, but the
results from this survey have not yet been produced.
To further investigate the project's impact on women's empowerment, we conducted an additional
analysis that separated the sample into Muslim and non-Muslim households. Interestingly, we found
that the project had a significant positive effect on women's autonomous income generation and their
decision making involvement for agricultural production and sales for the non-Muslim sub-sample.
This suggests that women from Muslim households in these areas may be facing barriers that the
project was unable to overcome. The project has not targeted these potential barriers, as the project's
theory of change assumed that these would not hinder the project's impact (see Section 1b), which is
an assumption that did not hold as these results suggest. Past research has found that, due to
patrilineal and patrilocal kinship systems and traditional gender norms, women face specific barriers
related to their freedom of movement and role in the household, as well as lack control over
economic resources such as assets and credit (Goetz and Sen Gupta, 1996; Donno and Russett, 2004;
Ambler et al., 2017). These findings underline the importance of a thorough understanding of the
general social context surrounding women's participation in the economy and the society in order to
provide improved support to women in similar settings through future projects.
46
Table 18: Impact of CCRIP on women's empowerment
2km from market 1km from market
ATE Obs ATE Obs
Household income from women's
activities (percentage points) 0.6 2,730 0.3 1,308
Women own enterprise (% likelihood) 0.4 2,730 -0.2 1,308
Women involved in decision making (joint or individual) for (% likelihood):
Household purchases 1.5 2,730 1.8 1,308
Children's education -0.2 2,730 -0.4 1,308
Crop or livestock prod. -1.6 1,894 1.3 859
Crop or livestock sales -4.7* 1,894 -2.1 859
b. Impact heterogeneity
Impact by geographic location
Table 19 presents CCRIP's impact on key indicators by geographic location, showing results for
households based in the Barisal, Patuakhali, Bhola and Barguna districts of the Barisal division (the
main area covered by the project), and for households located in the Khulna and Dhaka divisions. The
districts of Khulna and Dhaka could not be analyzed separately due to the smaller sample sizes for these
divisions.
We find that CCRIP's impact varied significantly across these areas. In terms of total income, there was
a significant impact of 60 per cent in Bhola district of Barisal, and a significant impact of 21 per cent in
the Dhaka division. In these areas we also find that households are significantly less likely to be poor
(by 18 and 13 per cent). However, the impact on total income in other areas was not significant.
These differences in impact on total income are shaped by varying impacts on the main income
components. For instance, in Barisal and Patuakhali districts, income from wage labour increased
significantly, as did income for those selling fish in Patuakhali, but a lack of impact in other income
components meant that there was no impact on total income. Income from wage labour significantly
increased also in Khulna, as did income for those selling fish or livestock, but a lack of impact on crop
income or income from household enterprises meant again that total income did not increase. For
Barguna, very large impacts were achieved for those selling crops, fish, and livestock, but a large
significant decrease in income from household enterprises curtailed the project's impact on total
income in this district. In Dhaka, the significant increase in total income, and reduction in poverty, was
driven by a large increase of 209 per cent in livestock income. For Bhola, where the largest impact on
total income was found, the impact was driven by large increases in income from fish sales and wage
labour.
Households in Barguna have the lowest average income at the district level, followed by households in
Khulna and Patuakhali, whilst households in Bhola, Barisal district, and Dhaka have relatively higher
47
incomes. Whilst we find significant increases in total income in Bhola and Dhaka, the promising
findings for Barguna suggest that factors other than wealth may have served to shape the income effect
across these geographic groups. For instance, the improvements in livestock income that drove the
income impact in Dhaka may have been due to an existing facilitating environment for livestock
marketing in this area. This is supported by secondary data showing that livestock ownership in Dhaka
is higher than in Khulna and Barisal (BBS, 2008). Dhaka is also located further inland, so may have
faced less vulnerability to climatic shocks compared to the coastal areas. In Barguna, better access to
markets seem to have pulled households away from household enterprise activities towards crop,
livestock and fish production and sales. The resulting reduction in income from eneterprises, however,
was not fully accounted for by improvements in income from these other sources. Finally for Khulna,
the poorest of the three divisions, there were clearly barriers to improving income from crop sales that
the project was unable to fully overcome, and which hindered the impact on total income as a result.
48
Table 19: Impact of CCRIP by districts of Barisal and Khulna and Dhaka divisions
Barisal Patuakhali Bhola Barguna Khulna districts Dhaka districts
ATE* Obs ATE Obs ATE Obs ATE Obs ATE Obs ATE Obs
Gross income per capita -9.1 427 -11.9 432 60.1*** 407 15.3 426 -1.7 454 21.0** 594
Above $1.90/ day pov. line (%
likelihood) -6.0 427 2.9 432 18.3*** 407 -3.8 426 -0.8 454 12.6** 594
Income from crop sale per ha. -58.8 334 34.2 286 15.4 338 363.7*** 320 -13.8 212 20.4 390
Income from fish sale -55.4 173 170.1** 214 158.7*** 298 124.9*** 373 164.1** 345 -88.4 123
Income from livestock sale -76.5 138 -110.7* 358 12.3 344 107.7** 395 120.8* 341 208.7*** 398
Income from waged labour per
capita 75.1*** 138 38.4*** 255 52.1*** 218 -5.5 194 26.6*** 304 3.3 182
Income from household
enterprise per capita -19.9 114 -25.5 130 61.6 142 -108.6*** 93 24.9 139 -24.6 229
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
ii. Impact by catchment area
Table 20 presents CCRIP's impact on key indicators for households located within 2km of the
CCRIP market but not within 1km of the connecting road, and for households who are located both
within 2km of the market and within 1km of the connecting road. Somewhat surprisingly, we find a
larger impact on total income for those not located close to the road (13 per cent), and a non-
significant impact for those located closer to the road. We also find that households that are only
within a market catchment area (i.e. further than 1km of a road) were less likely to be poor. These
impacts on those located further from the road were driven by a larger impact on income from
selling crops, which averaged 150 per cent compared to 81 percent for those located closer to the
road, and a 69 per cent increase in income from livestock sales. Housheolds located closer to the
road observed a larger increase in income from fish sales and wage labour, but no significant
increase in total income.
Households who are located close to both the CCRIP market and the connecting road were
theorectically expected to benefit the most from the project as they were receiving two types of
support. However, our results show that those located further from the connecting road benefitted
more in terms of total income. Given that households located further from the road are poorer (see
Section 3b), it may be the case that within the market catchment area, the impact of the project was
more pro-poor. Unlike previous studies of infrastructure investments that found a lack of impact on
the poorest households, in the case of CCRIP poorer households starting from a lower base of
income, and facing more barriers, seemingly experienced larger increases thanks to the project's
support (Khandker and Koolwall, 2010). In addition, these results also suggest that the connecting
roads were perhaps more important for certain livelihood activities, such as the fish sales and wage
labour, than for the sale of crops or livestock products.
Table 20: Impact of CCRIP by catchment area
Market only Market and road
ATE* Obs ATE Obs
Gross income per capita 12.7* 1,491 7.5 1,249
Above $1.90/ day pov. line
(% likelihood) 5.3* 1,491 3.8 1,249
Income from crop sales 150.4*** 985 81.2** 891
Income from fish sales 59.6* 865 70.7** 661
Income from livestock sales 69.3** 1,170 -32.9 1,001
Income from wage labour
per capita 17.0 768 18.4** 523
Income from household
enterprise per capita -26.6 457 5.7 390
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
iii. Impact by livelihood activities
The whole sample used for this analysis includes households that are primarily farm households, and
those that rely primarily on other income generating activities. In order to assess whether CCRIP
impact varied by these two household groups, Table 21 presents impacts on the income of
households who were involved in farming (those who sold any crops, produced any fish, or kept any
livestock during the 12 month study period), and for those households not involved in farming. We
find that the project's impact is driven by improvements amongst farming households, with their total
income increasing significantly by 16 per cent, compared to a negligible impact for non-farming
households. Along with increased farm income, the impact on total income for farming households
was also driven by higher income for wage labourers. As a result of these impacts, farming
households were 6 per cent less likely to be poor, while no impact was found for non-farming
households.
The average income for farming households is around 30 per cent lower than the average income of
non-farming households, and farming households also are more likely to be below the poverty line.
The fact that the project only impacted farming households provides further evidence that the project
was more beneficial for poorer households within the market catchment area. It should also be
considered, however, that project activities may not have been suited to the livelihood needs of non-
farming households in the project areas, with markets seemingly being beneficial for sales of crops,
fish, and in rare cases livestock, but not for providing opportunities for household enterprises and, in
some cases, wage labour. This insight is also supported by the income impacts for households
located within the 1km catchment area (see Section 4a-iv).
Table 21: Impact of CCRIP by livelihood activities
Farm households Non-farm households
ATE* Obs ATE Obs
Gross income per capita 15.9*** 2,163 -6.7 577
Above $1.90/ day pov. line
(% likelihood) 5.5** 2,163 -0.9 577
Wage income per capita 14.2** 991 14.03 300
Household enterprise
income per capita 3.0 648 -20.9 199
5. Conclusion
The Coastal Climate Resilient Infrastructure Project aimed to improve the livelihoods of poor and
remote households in Southwest Bangladesh through improved markets and roads. In particular, the
project aimed to improve the access and useability of community markets during seasonal flooding.
Using data from an in-depth household questionnaire combined with qualitative interviews, we have
rigorously assessed the project's impact on a range of impact indicators relating to income; crop, fish
and livestock production and sales; assets, food security and education; financial inclusion; and
women's empowerment. We have assessed impact on the whole sample, as well as for a range of
sub-groups, including by geographic location, location within the market catchment area, and by
livelihood activity, integrating findings from the qualitative data to help to explain the mechanisms
that shaped the project's impact.
Overall, we have found that this type of focused infrastructure project can improve the climate
resilience and accessibility of local markets, leading to significant improvements in income, assets
and food security. This is achieved by improving households' market access in both the dry and
monsoon seasons, and thus the marketing practices of beneficiaries, particularly in terms of crop and
fish sales. As well as infrastructure support, we find that these impacts were facilitated through
capacity building support to the local institutions managing the markets, which contributed to the
sustainable management of the project markets. Through sub-group analyses we find that this type of
project can be particularly beneficial for poorer, more remote households, although some impacts,
such as on livestock income and wage labour vary depending on facilitating conditions.
A number of lessons can be drawn from these findings to improve the impact of future climate
resilient infrastructure projects. First, future projects should pay special attention to ensuring
households have access to high-quality agricultural inputs, as improved access to output markets
does not necessarily improve access to inputs especially for capital constrained households.. In
addition to improving input access, agricultural productivity and income could also be improved by
providing complementary training and technology support.
Second, future projects should consider different components of beneficiaries' livelihoods and
provide activities to stimulate the main sources of income, which may vary for different areas. In
some cases for CCRIP, there was a lack of impact on the main income sources for some households
(mainly wage labour and household enterprises), leading to a lack of impact on total income. Impact
on income could thus be enhanced by offering complementary support for the livelihood activities
that are the most important in each local context. In the case of wage labour, this could involve a
redesign of the LCS activities to ensure that the valuable employment and training provided does
not remain short-term in nature and includes support to establish linkages with local labor market
for sustained impacts. In terms of household enterprises, these activities could be improved by
facilitating easier entry into local markets for small shops and traders, as well as providing credit and
training for setting up and managing these businesses.
Finally, future projects should provide more extensive support to improve the income generating
opportunities and overall empowerment of women, especially in countries such as Bangladesh where
women face ingrained barriers to their mobililty and autonomy. Although the LCS impacts for
CCRIP were promising, these activities combined with market and road support that did not
explicity target these barriers were largely insufficient for improving women's empowerment,
particularly in Muslim households. Based on the success of initiatives such as BRAC's
Empowerment and Livelihood for Adolescents program in Bangladesh (as well as Afghanistan,
Haiti, Sierra Leone, South Sudan, Tanzania and Uganda - see Bandiera et al. 2018) future projects
could provide multi-faceted support to improve the hard and soft skills of women, provided within a
safe space environment, and involve the wider society to ensure sustainability of impacts.
References
Ahmed, A.U., Ahmad, K., Chou, V., Hernandez, R., Menon, P., Naeem, F., Naher, F., Quabili, W.,
Sraboni, E. and Yu, B. 2013. The status of food security in the Feed the Future Zone and other
regions of Bangladesh. Paper Prepared for USAID Grant No. EEM-G-00-04-00013-00. Dhaka,
Bangladesh: IFPRI.
Ahmed, Z. 2010. Role of local government in indigienous market management in the rural areas of
Bangladesh: Do these markets play development roles? Journal of Sustainable Development, 3(1):
120-135
Alkire, S. and M.E. Santos. 2010. Acute multidimensional poverty: A new index for developing
countries. Oxford Poverty and Human Development Initiative (OPHI) Working Paper No. 38.
Oxford, UK: OPHI. https://www.econstor.eu/bitstream/10419/48297/1/3_alkire.pdf
Anderson, J., Moler, A. and Kretchun, N. 2016. National survey and segmentation of smallholder
households in Bangladesh: Understanding their demand for financial, agricultural, and digital
solutions. CGAP Working Paper. Washington, D.C., USA: CGAP.
Arslan, A., Cavatassi, R., Alfani, F., McCarthy, N., Lipper, L., and Kokwe, M. 2018a.
Diversification Under Climate Variability as Part of a CSA Strategy in Rural Zambia, The Journal of
Development Studies, 54 (3): 457-480.
Arslan, A., Higgins, D. & Winters, P. 2018b. Impact Assessment Report: Integrated Rice Production
Enhancement Project (IRPEP), The Philippines. Rome, Italy: IFAD. Available at:
https://drive.google.com/file/d/1VRVKDvdkJedsrE9-ncH1HiLUgGS7nmLI/view?usp=sharing
Austin, P. C. 2009. Balance diagnostics for comparing the distribution of baseline covariates
between treatment groups in propensity-score matched samples. Statistics in Medicine, 28(25): 3083-
3107
Austin, P.C. 2011. An introduction to propensity score methods for reducing the effects of
confounding in observational studies. Multivariate Behavioural Research, 46(3): 399-424
Austin, P.C. and E.A. Stuart. 2015. Moving towards best practice when using inverse probability of
treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in
observational studies. Statistics in Medicine, 34: 3661-3679
Ballard, T., Kepple, A.W. and Cafiero, C. 2013. The Food Insecurity Experience Scale:
Development of a global standard for monitoring hunger worldwide. FAO Technical Paper. Rome,
Italy: FAO.
Bandiera, O., N. Buehren, R. Burgess, M. Goldstein, S. Gulesci, I. Rasul and M. Sulaiman. 2018.
Women’s Empowerment in Action: Evidence from a Randomized Control Trial in Africa. London:
Economic Organization and Public Policy
Barrett, C. 2008. Smallholder market participation: Concepts and evidence from eastern and
southern Africa. Food Policy, 33: 299-317
Baser, O. 2006. Too much ado about Propensity Score Models? Comparing methods of Propensity
Score Matching. Value in Health 9(6): 377-385.
Bokelmann, W. And M.E. Adamseged. 2016. Contributing to a better understanding of the value
chain framework in developing countries. Invited poster presented at the 5th International
Conference of the African Association of Agricultural Economists, September 23-26, 2016, Addis
Ababa, Ethiopia.
Caliendo, M. and S. Kopeinig. 2008. Some practical guideance for the implementation of propensity
score matching. Journal of economic surveys, 22(1):31-72
Carletto, G. Covarrubias, K. Davis, B., Krausova, M. and Winters, P. 2007. Rural Income
Generating Activities Study: Methodological note on the construction of income aggregates.Rome,
Italy: FAO.
Carletto, G., Ruel, M., Winters, P. & Zezza, A. 2015. Farm-level pathways to improved nutritional
status: Introduction to the special issue. Journal of Development Studies, 51: 945-957.
Cropper, M. and Griffiths, C. 1994. The interaction of population growth and environmental quality,
The American Economic Review, 84(2): 250-254.
Dehejia, R.H. and Wahba, S. 1999. Causal effects in nonexperimental studies: Reevaluating the
evaluation of training programs. Journal of the American Statistical Association, 94(448): 1053-
1062.
Donno, D. and Russett, B. 2004. Islam, authoritarianism, and female empowerment: What are the
linkages? World Politics, 56(4): 582-607.
Ellis, F. 1999. Rural livelihood diversity in developing countries: Evidence and policy implications.
ODI Natural Resource Perspectives No. 40. https://www.odi.org/sites/odi.org.uk/files/odi-
assets/publications-opinions-files/2881.pdf
Fan, M., J. Shen, L. Yuan, R Jiang, X. Chen, W.J. Davies, F. Zhang. 2012. Improving crop
productivity and resource use efficiency to ensure food security and environmental quality in China.
Journal of Experimental Botany, 63(1): 13-24
FAO (Food and Agriculture Organisation of the United Nations). 2003. Planning and designing
rural markets. Rome, Italy: FAO.
Ferreira, F.H.G., Chen, S., Dabalen, A.L., Dikhanov, Y.M., Hamadeh, N., Jolliffe, D.M., Narayan,
A., Prydz, E.B., Revenga, A.L., Sangraula, P., Serajuddin, U., Yoshida, N. 2015. A global count of
the extreme poor in 2012: Data issues, methodology and initial results. Policy Research Paper No
WPS 7432. Washington, DC., USA: World Bank
Filmer, D. and Pritchett, L.H. 2001. Estimating wealth effects without expenditure data – or tears:
An application to educational enrolments in states of India. Demography, 38(1): 115-132
Gallo, A. 2016. A refresher on statistical significance. Harvard Business Review.
https://hbr.org/2016/02/a-refresher-on-statistical-significance
Garbero, A. 2016. Measuring IFAD's impact: Background paper to the IFAD9 Impact Assessment
Initiative. IFAD Research Series Issue 7. Rome, Italy: IFAD
Goetz, A.M. and Sen Gupta, R. 1996. Who takes the credit? Gender, power, and control over loan
use in rural credit programs in Bangladesh. World Development, 24(1): 45-63.
Gulati, A., N. Minot, C. Delgado, and S. Bora. 2005. Growth in high-value agriculture in Asia and
the emergence of vertical links with farmers. Paper presented at the workshop "Linking small-scale
producers to markets: Old and new challenges" The Work Bank, 15 December 2005.
Hetherington, J.B., Wiethoelter, A.K., Negin, J. & Mor, S.M. 2017. Livestock ownership, animal
source foods and child nutritional outcomes in seven rural village clusters in Sub-Saharan Africa.
Agriculture and Food Security, 6(9): 1-11.
Huq, N., J. Huge, E. Boon, and A.K. Gain. 2015. Climate change impacts in agricultural
communities in rural areas of coastal Bangladesh: A tale of many stories. Sustainability, 7:8437-
8460
Islam, M.S., Sultana, S., Saifunnahar, M. and Miah, M.A. 2014. Adaptation of char livelihood in
flood and river erosion areas through indigenous practice: A study on Bhuapur Riverine Area in
Tangail. Journal of Environmental Science and Natural Resources, 7(1): 13-19.
Jahnke, H.E. 1982. Livestock production systems and livestock development in tropical Africa. Kiel,
Germany: Kieler Wissenschaftersverlag Vauk
Jones, L., Jaspars, S., Pavanello, S., Ludi, E., Slater, R., Arnall, A., Grist, N. and Mtisi, S. 2010.
Responding to a changing climate: Exploring how disaster risk reduction, social protection and
livelihoods approaches promote features of adaptive capacity. ODI Working Paper No. 319.
London, UK: Overseas Development Institute.
Khandker, R., Z. Bakht, and G.B. Koolwal. 2009. The poverty impact of rural roads: Evidence from
Bangladesh. Economic Development and Cultural Change, Vol 57(4): 685-722
Khandker, S. R., Koolwal, G. B. & Samad, H. A. 2010. Handbook on impact evaluation:
Quantitative methods and practice. Washington, DC, USA: World Bank.
Kirtman, B., S.B. Power, J.A. Adedoyin, G.J. Boer, R. Bojariu, I. Camilloni, F.J. Doblas-Reyes,
A.M. Fiore, M. Kimoto, G.A. Meehl, M. Prather, A. Sarr, C. Schär, R. Sutton, G.J. van Oldenborgh,
G. Vecchi and H.J. Wang. 2013: Near-term Climate Change: Projections and Predictability. In:
Stocker, T.F., D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex
and P.M. Midgley. (eds.) Climate Change 2013: The Physical Science Basis. Contribution of
Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change.
Cambridge, UK: Cambridge University Press.
Meybeck, A., J. Lankoski, S, Redfern, N, Azzu, and V. Gitz. 2012. Building resilience for
adaptation to climate change in the agriculture sector. Proceedings of a join FAO/OECD workshop
23-24 April 2012. Rome, Italy: FAP
Nishat, A., N. Mukherjee, E. Roberts, and A. Hasemann. 2013. A range of approaches to address
loss and damage from climate change impacts in Bangladesh. Loss and Damage Publication. Bonn,
Germany: Germanwatch
Rahman, S. and M.A. Rahman. 2015. Climate extremes and challenges to infrastructure development
in coastal cities in Bangladesh. Weather and Climate Extremes, 7:96-108
Rehima, M., K. Belay, A. Dawit, S. Rashid. 2013. Factors influencing farmers' crop diversification:
Evidence from SNNPR, Ethiopia. International Journal of Agricultural Studies, 3(8): 558-565
Relief Web. 2008. Cyclone Sidr in Bangladesh: Damage, loss and needs assessment for disaster
recovery and reconstruction. A report prepared by the Government of Bangladesh assisted by the
international development community with financial support from the European Commission. New
York City, USA: Relief Web
Relief Web. 2009. Cyclone Aila losses in Bangladesh estimated at USD269 million.
https://reliefweb.int/report/bangladesh/cyclone-aila-losses-bangladesh-estimated-269-mln-usd
Roy, S.K., F Bilkes, K. Islam, G. Ara, P. Tanner, I. Wosk, A.S. Rahman, B. Chakraborty, S.P. Jolly,
W. Khatun. 2008. Impact of pilot project of Rural Maintenance Programme (RMP) on destitute
women: CARE, Bangladesh. Food and Nutrition Bulletin, 29(1): 67-75
Saha, C.K. 2014. Dynamics of disaster-induced risk in southwestern coastal Bangladesh: An analysis
on Tropical Cyclone Aila 2009. Natural Hazards. 75(1): 727-754
Sarker, M.H., Huque, I., Alam, M. and Koudstaal, R. 2003. Rivers, chars and char dwellers of
Bangladesh. International Journal of River Basin Management, 1(1): 61-80.
Sheoran, J. 2016. Aspects of women empowerment: A brief overview. International Research
Journal of Management, Science and Technology, 7(4):135-140
Svensson, J. and D.Y. Drott. 2010. Tuning in the market signal: The impact of market price
information on agricultural outcomes. Institute for International Economic Studies Working Paper.
Stockholm, Sweden: Stockholm University
Vallejo, L. and M. Mullan. 2017. Climate-resilient infrastructure: Getting the policies right. OECD
Environmental Working Papers No. 121. Paris, France: OECD Publishing
White, H. Theory-based impact evaluation: Principles and practice. Journal of Development
Effectiveness, 1(3): 271-284.
Wooldridge, J.M. 2010. Econometric analysis of cross section and panel data. 2nd
ed. Cambridge,
MA: MIT Press
World Bank. 2001. Engendering Development: A World Bank Policy Research Report. Washington,
DC., USA: The World Bank
World Bank, 2017. Atlas of sustainable development goals 2017: From World Development
Indicators. Washington, DC, USA: World Bank
World Bank, 2017. GDP per capita (current US$), Bangladesh. [Online] Available at:
https://data.worldbank.org/indicator/NY.GDP.PCAP.CD
Appendix I: Mean values for impact indicators
I.A: Treatment and control means for impact indicators of agricultural productivity and sales
Treatment Control Diff.
Mean Obs Mean Obs
Value of ag. production per ha (BDT):
Full year 149,466 806 43,160 1,070 106,306*
Dry seasons only 59,452 806 39,944 1,070 19,508***
Monsoon season only 38,666 806 41,020 1,070 -2,354
Gross margins per ha (BDT). 111,945 806 10,632 1,070 101,313*
Proportion of harvest sold (%) 33.2 806 27.1 1,070 6.1***
Sold at market (%) 50.3 586 41.6 683 8.7***
Income from crop sale per ha. (BDT):
Full year 40,879 806 22,455 1,070 18,424***
Dry seasons 31,577 806 17,267 1,070 14,310***
Monsoon season 48,351 806 26,926 1,070 21,425***
Nr. crop varieties (count) 2.6 806 2.6 1,070 0
Grew at least one high-value crop
(%) 43.9 806 38.3 1,070 5.6**
Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at
the 10, 5 and 1% levels, respectively.
I.B: Treatment and control means for impact indicators of agricultural input use
Treatment Control Diff.
Mean Obs Mean Obs
Land cultivated (ha.) 3.3 806 2.8 1,070 0.5*
Value of inputs per ha. (BDT) 37,522 806 32,528 1,070 4,994**
Seeds 6,457 806 3,606 1,070 2,851**
Fertiliser 4,653 806 4,054 1,070 599
Crop Protection† 3,519 806 3,717 1,070 -198
Labour 27,543 806 25,585 1,,070 1,958
Input productivity of (BDT):
Seeds 66.9 710 40.4 900 26.5
Fertiliser 774.3 671 29.5 854 744.8
Crop Protection 25,970 629 119.6 780 25,850
Labour 329.6 806 201.9 1,070 127.7*
Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at
the 10, 5 and 1% levels, respectively.
I.C: Treatment and control means for impact indicators of fish production and sale
Treatment Control Diff.
Mean Obs Mean Obs
Gross value of fish production
(BDT) 33,181 626 26,746 900 6,435
Net value of fish inputs (BDT) 241,260 626 235,708 900 5,552
Revenue from fish sale (BDT) 18,573 626 5,429 900 13,144
Sold fish at market (%) 55.0 129 41.3 143 13.7**
Value of fish consumption per
capita (BDT) 2,390 626 2,201 900 189
Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at
the 10, 5 and 1% levels, respectively.
I.D: Treatment and control means for impact indicators of livestock rearing and sale
Treatment Control Diff.
Mean Obs Mean Obs
Gross value of livestock
production (BDT) 40,477 944 31,490 1,227 8,987
Net value of livestock production
(BDT) 410,993 944 404,329 1,227 6,664
Revenue from livestock and
product sale (BDT) 30,734 944 22,811 1,227 7,923
Sold livestock or livestock
product (%) 52.4 944 50.3 1,227 2.1
Milk productivity per cow (BDT) 8,342 356 6,218 532 2,124***
Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at
the 10, 5 and 1% levels, respectively.
I.E: Treatment and control means for impact indicators of income and livelihood composition
Treatment Control Diff.
Mean Obs Mean Obs
Gross income per capita (BDT) 79,588 1,188 69,177 1,552 10,411
Net income per capita (BDT) 55,658 1,188 45,277 1,552 10,381*
Income from waged labour per
capita (BDT) 51,328 554 35,901 737 15,427
Income from household enterprise
per capita (BDT) 98,508 369 98,628 478 -120
Nr. income sources (count) 7.9 1,188 7.6 1,552 0.3**
Above $1.90/ day pov. line (%) 75.3 1,188 73.3 1,552 2.0
Proportion of income from (%):
Crop sale 16.1 1,188 13.3 1,552 2.8***
Fish sale 2.1 1,188 1.3 1,552 0.8**
Livestock sale 6.2 1,188 6.6 1,552 -0.4
Formal waged labour 22.2 1,188 20.5 1,552 1.7
Casual waged labour 10.0 1,188 11.8 1,552 -1.8*
Household enterprises 23.5 1,188 23.4 1,552 0.1
Remittances 12.7 1,188 16.2 1,552 -3.5***
Land rental 1.1 1,188 1.2 1,552 -0.1
Other sources 4.1 1,188 4.1 1,552 0
Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at
the 10, 5 and 1% levels, respectively.
I.F: Treatment and control means for impact indicators of assets, food security and education
Treatment Control Diff.
Mean Obs Mean Obs
Asset indices:
Household durable assets 1.2 1,188 1.2 1,552 0
Productive assets 0.6 1,188 0.6 1,552 0
TLU 0.8 1,188 0.8 1,552 0
FIES score 2.2 1,188 2.6 1,552 -0.4***
Worried about having enough
food (%) 51.7 1,188 62.8 1,552 -11.1***
Dietary Diversity Score 10.5 1,188 10.6 1,552 -0.1
School-age children enrolled (%) 70.0 952 71.1 1,252 -1.1
Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at
the 10, 5 and 1% levels, respectively.
I.G: Treatment and control means for impact indicators of financial inclusion
Treatment Control Diff.
Mean Obs Mean Obs
Have formal bank account (%) 20.6 1,188 17.8 1,552 2.8*
Have other account
(MFI, NBFI, mobile money) (%) 19.6 1,188 24.4 1,552 -4.8***
Took a loan in past year (%) 51.0 1,188 52.5 1,552 -1.5
Savings per capita (BDT) 2,060 1,188 2,018 1,552 42
Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at
the 10, 5 and 1% levels, respectively.
I.H: Treatment and control means for impact indicators of women's empowernment
Treatment Control Diff.
Mean Obs Mean Obs
Household income from women's
activities (% of total income) 1.6 1,182 1.3 1,548 0.3
Women own enterprise (%) 2.1 1,182 2.0 1,548 0.1
Women involved in decision making (joint or individual) for (%):
Household purchases 39.8 1,182 37.7 1,548 2.1
Children's education 82.9 1,182 85.7 1,548 -2.8**
Crop or livestock prod. 57.8 778 61.1 1,043 -3.3
Crop or livestock sales 52.3 778 57.2 1,043 -4.9**
Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at
the 10, 5 and 1% levels, respectively.
Appendix II: Distribution of Propensity Scores before and after trimming
Before trimming
After trimming
Appendix III: Results from the secondary IPWRA model
III.A: Impact of CCRIP on agricultural productivity and sales
2km from market 1km from market
ATE* Obs ATE Obs
Value of ag. production per ha:
Full year 15.8 1,876 14.8 866
Dry seasons only 148.7** 1,876 122.4** 866
Monsoon season only -24.4 1,876 -4.4 866
Gross margins per ha. 6.0 1,876 5.6 866
Proportion of harvest sold 6.9** 1,876 3.8 866
Sold at market (% likelihood) 6.7 1,269 15.3** 598
Income from crop sales per ha.:
Full year 114.3** 1,876 134.1*** 866
Dry seasons 186.0** 1,876 143.8** 866
Monsoon season 99.2 1,876 79.8 866
Nr. crop varieties (count) 3.1 1,876 -2.3 866
Grew at least one high-value
crop (% likelihood) 4.9
1,876 3.0
866
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
III.B: Impact of CCRIP on agricultural input use
2km from market 1km from market
ATE* Obs ATE Obs
Land cultivated (ha.) -6.8 1,876 76.2 866
Value of inputs per ha. 13.5 1,876 24.4 866
Seeds 9.2 1,876 21.5 866
Fertiliser 5.9 1,876 68.6* 866
Crop Protection† 32.9 1,876 96.0** 866
Labour 15.9 1,876 47.2 866
Input productivity of:
Seeds 7.0 1,610 26.6** 750
Fertiliser 8.3 1,525 -1.4 708
Crop Protection 3.6 1,409 0.8 636
Labour 6.6 1,876 -4.0 866
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
† Crop protection includes following inputs: pesticides, herbicides, fungicides, insecticide
III.C: Impact of CCRIP on fish production and sale
2km from market 1km from market
ATE* Obs ATE Obs
Gross value of fish production -18.1 1,456 2.0 664
Net value of fish inputs -1.3 1,456 -1.9 664
Revenue from fish sales 51.0 1,456 47.0 664
Sold fish at market (% likelihood) 8.6 257 -1.24 101
Value of fish consumption per
capita -14.3 1,456 -17.2 664
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
III.D: Impact of CCRIP on livestock rearing and sale
2km from market 1km from market
ATE* Obs ATE Obs
Gross value of livestock production 9.1 2,171 14.6 1,006
Net value of livestock production -0.7 2,171 1.2 1,006
Revenue from livestock and product sales 22.3 2,171 2.7 1,006
Sold livestock or livestock product (%
likelihood) 2.5 2,171 0.1 1,006
Milk productivity per cow -31.1 2,267 16.0 404
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
III.E: Impact of CCRIP on income and livelihood composition
2km from market 1km from market
ATE* Obs ATE Obs
Gross income per capita 6.0 2,740 7.2 1,312
Net income per capita -0.7 2,740 -2.4 1,312
Income from wage labour
per capita 14.8** 1,291 10.5 607
Income from household
enterprise per capita -4.9 847 8.3 423
Nr. income sources (count) 0.2 2,740 0.4 1,312
Above $1.90/ day pov. line
(% likelihood) 2.6 2,740 5.5 1,312
Proportion of income from (percentage points):
Crop sales 0.6 2,740 5.2** 1,312
Fish sales 0.6 2,740 0.7 1,312
Livestock sales -0.9 2,740 -1.7 1,312
Formal wage labour 2.4 2,740 2.0 1,312
Casual wage labour -1.1 2,740 -2.5 1,312
Household enterprises -0.6 2,740 -3.9 1,312
Remittances -1.4 2,740 -0.6 1,312
Land rental -0.1 2,740 -0.4 1,312
Other sources -0.2 2,740 0.9 1,312
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
III.F: Impact of CCRIP on indicators of assets, food security and education
2km from market 1km from market
ATE Obs ATE Obs
Asset indices:
Household durable
assets 0.1 2,740 15.1** 1,312
Productive assets -0.01 2,740 6.3** 1,312
TLU -0.1 2,740 -12.6 1,312
FIES score -13.8%** 2,740 -23.3%*** 1,312
Worried about having
enough food (% likelihood) -11.2*** 2,740 -17.5*** 1,312
Dietary Diversity Score -0.6 2,740 -39.5 1,312
School-age children enrolled
(percentage points) -1.7 2,204 3.4 1,056
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
III.G: Impact of CCRIP on financial inclusion
2km from market 1km from market
ATE* Obs ATE Obs
Have formal bank account
(% likelihood) 2.2 2,740 2.4 1,312
Have other account
(MFI, NBFI, mobile money)
(% likelihood) -4.2* 2,740 0.2 1,312
Took a loan in past year (%
likelihood) 0.2 2,740 -3.9 1,312
Savings per capita -4.2 2,740 5.1 1,312
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.
*Coefficients represent percentage change unless specified otherwise.
III.H: Impact of CCRIP on women's empowernment
2km from market 1km from market
ATE Obs ATE Obs
Household income from women's
activities (percentage points) 0.3 2,730 0.5 1,308
Women own enterprise (% likelihood) 0.2 2,730 1.1 1,308
Women involved in decision making (joint or individual) for (% likelihood):
Household purchases 3.1 1,821 6.9 1,308
Children's education -1.0 1,821 3.1 1,308
Crop or livestock prod. -0.3 1,821 3.5 838
Crop or livestock sales -2.3 1,821 -0.5 838
Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.